basic.py 199 KB
Newer Older
wxchan's avatar
wxchan committed
1
# coding: utf-8
2
"""Wrapper for C API of LightGBM."""
3

4
import abc
wxchan's avatar
wxchan committed
5
import ctypes
6
import inspect
7
import json
wxchan's avatar
wxchan committed
8
import warnings
9
from collections import OrderedDict
10
from copy import deepcopy
11
from enum import Enum
12
from functools import wraps
13
from os import SEEK_END, environ
14
15
from os.path import getsize
from pathlib import Path
16
from tempfile import NamedTemporaryFile
17
from typing import TYPE_CHECKING, Any, Callable, Dict, Iterable, Iterator, List, Optional, Set, Tuple, Union
wxchan's avatar
wxchan committed
18
19
20
21

import numpy as np
import scipy.sparse

22
23
24
25
from .compat import (
    PANDAS_INSTALLED,
    PYARROW_INSTALLED,
    arrow_cffi,
26
    arrow_is_boolean,
27
28
29
30
31
32
33
34
35
36
37
38
39
    arrow_is_floating,
    arrow_is_integer,
    concat,
    dt_DataTable,
    pa_Array,
    pa_chunked_array,
    pa_ChunkedArray,
    pa_compute,
    pa_Table,
    pd_CategoricalDtype,
    pd_DataFrame,
    pd_Series,
)
wxchan's avatar
wxchan committed
40
41
from .libpath import find_lib_path

42
43
44
if TYPE_CHECKING:
    from typing import Literal

45
46
47
48
49
50
51
    # typing.TypeGuard was only introduced in Python 3.10
    try:
        from typing import TypeGuard
    except ImportError:
        from typing_extensions import TypeGuard


52
__all__ = [
53
54
55
56
57
58
    "Booster",
    "Dataset",
    "LGBMDeprecationWarning",
    "LightGBMError",
    "register_logger",
    "Sequence",
59
60
]

61
_BoosterHandle = ctypes.c_void_p
62
_DatasetHandle = ctypes.c_void_p
63
64
_ctypes_int_ptr = Union[
    "ctypes._Pointer[ctypes.c_int32]",
65
    "ctypes._Pointer[ctypes.c_int64]",
66
]
67
68
_ctypes_int_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_int32]]",
69
    "ctypes.Array[ctypes._Pointer[ctypes.c_int64]]",
70
]
71
72
_ctypes_float_ptr = Union[
    "ctypes._Pointer[ctypes.c_float]",
73
    "ctypes._Pointer[ctypes.c_double]",
74
75
76
]
_ctypes_float_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_float]]",
77
    "ctypes.Array[ctypes._Pointer[ctypes.c_double]]",
78
]
79
_LGBM_EvalFunctionResultType = Tuple[str, float, bool]
80
_LGBM_BoosterBestScoreType = Dict[str, Dict[str, float]]
81
_LGBM_BoosterEvalMethodResultType = Tuple[str, str, float, bool]
82
_LGBM_BoosterEvalMethodResultWithStandardDeviationType = Tuple[str, str, float, bool, float]
83
84
_LGBM_CategoricalFeatureConfiguration = Union[List[str], List[int], "Literal['auto']"]
_LGBM_FeatureNameConfiguration = Union[List[str], "Literal['auto']"]
85
86
87
88
_LGBM_GroupType = Union[
    List[float],
    List[int],
    np.ndarray,
89
90
91
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
92
]
93
94
_LGBM_PositionType = Union[
    np.ndarray,
95
    pd_Series,
96
]
97
98
99
100
101
102
_LGBM_InitScoreType = Union[
    List[float],
    List[List[float]],
    np.ndarray,
    pd_Series,
    pd_DataFrame,
103
104
105
    pa_Table,
    pa_Array,
    pa_ChunkedArray,
106
]
107
108
109
110
111
112
113
114
115
_LGBM_TrainDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
    scipy.sparse.spmatrix,
    "Sequence",
    List["Sequence"],
116
    List[np.ndarray],
117
    pa_Table,
118
]
119
_LGBM_LabelType = Union[
120
121
    List[float],
    List[int],
122
123
    np.ndarray,
    pd_Series,
124
125
126
    pd_DataFrame,
    pa_Array,
    pa_ChunkedArray,
127
]
128
129
130
131
132
133
_LGBM_PredictDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
134
135
    scipy.sparse.spmatrix,
    pa_Table,
136
]
137
138
139
140
_LGBM_WeightType = Union[
    List[float],
    List[int],
    np.ndarray,
141
142
143
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
144
]
145
146
147
148
149
150
151
152
153
154
155
156
157
_LGBM_SetFieldType = Union[
    List[List[float]],
    List[List[int]],
    List[float],
    List[int],
    np.ndarray,
    pd_Series,
    pd_DataFrame,
    pa_Table,
    pa_Array,
    pa_ChunkedArray,
]

158
159
160
ZERO_THRESHOLD = 1e-35


161
162
163
164
def _is_zero(x: float) -> bool:
    return -ZERO_THRESHOLD <= x <= ZERO_THRESHOLD


165
def _get_sample_count(total_nrow: int, params: str) -> int:
166
    sample_cnt = ctypes.c_int(0)
167
168
169
170
171
172
173
    _safe_call(
        _LIB.LGBM_GetSampleCount(
            ctypes.c_int32(total_nrow),
            _c_str(params),
            ctypes.byref(sample_cnt),
        )
    )
174
175
    return sample_cnt.value

wxchan's avatar
wxchan committed
176

177
class _MissingType(Enum):
178
179
180
    NONE = "None"
    NAN = "NaN"
    ZERO = "Zero"
181
182


183
class _DummyLogger:
184
    def info(self, msg: str) -> None:
185
        print(msg)  # noqa: T201
186

187
    def warning(self, msg: str) -> None:
188
189
190
        warnings.warn(msg, stacklevel=3)


191
192
193
_LOGGER: Any = _DummyLogger()
_INFO_METHOD_NAME = "info"
_WARNING_METHOD_NAME = "warning"
194
195


196
197
198
199
def _has_method(logger: Any, method_name: str) -> bool:
    return callable(getattr(logger, method_name, None))


200
def register_logger(
201
202
203
    logger: Any,
    info_method_name: str = "info",
    warning_method_name: str = "warning",
204
) -> None:
205
206
207
208
    """Register custom logger.

    Parameters
    ----------
209
    logger : Any
210
        Custom logger.
211
212
213
214
    info_method_name : str, optional (default="info")
        Method used to log info messages.
    warning_method_name : str, optional (default="warning")
        Method used to log warning messages.
215
    """
216
    if not _has_method(logger, info_method_name) or not _has_method(logger, warning_method_name):
217
        raise TypeError(f"Logger must provide '{info_method_name}' and '{warning_method_name}' method")
218
219

    global _LOGGER, _INFO_METHOD_NAME, _WARNING_METHOD_NAME
220
    _LOGGER = logger
221
222
    _INFO_METHOD_NAME = info_method_name
    _WARNING_METHOD_NAME = warning_method_name
223
224


225
def _normalize_native_string(func: Callable[[str], None]) -> Callable[[str], None]:
226
    """Join log messages from native library which come by chunks."""
227
    msg_normalized: List[str] = []
228
229

    @wraps(func)
230
    def wrapper(msg: str) -> None:
231
        nonlocal msg_normalized
232
233
        if msg.strip() == "":
            msg = "".join(msg_normalized)
234
235
236
237
238
239
240
241
            msg_normalized = []
            return func(msg)
        else:
            msg_normalized.append(msg)

    return wrapper


242
def _log_info(msg: str) -> None:
243
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
244
245


246
def _log_warning(msg: str) -> None:
247
    getattr(_LOGGER, _WARNING_METHOD_NAME)(msg)
248
249
250


@_normalize_native_string
251
def _log_native(msg: str) -> None:
252
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
253
254


255
def _log_callback(msg: bytes) -> None:
256
    """Redirect logs from native library into Python."""
257
    _log_native(str(msg.decode("utf-8")))
258
259


260
def _load_lib() -> ctypes.CDLL:
261
    """Load LightGBM library."""
wxchan's avatar
wxchan committed
262
263
264
    lib_path = find_lib_path()
    lib = ctypes.cdll.LoadLibrary(lib_path[0])
    lib.LGBM_GetLastError.restype = ctypes.c_char_p
265
    callback = ctypes.CFUNCTYPE(None, ctypes.c_char_p)
266
    lib.callback = callback(_log_callback)  # type: ignore[attr-defined]
267
    if lib.LGBM_RegisterLogCallback(lib.callback) != 0:
268
        raise LightGBMError(lib.LGBM_GetLastError().decode("utf-8"))
wxchan's avatar
wxchan committed
269
270
    return lib

wxchan's avatar
wxchan committed
271

272
273
# we don't need lib_lightgbm while building docs
_LIB: ctypes.CDLL
274
if environ.get("LIGHTGBM_BUILD_DOC", False):
275
    from unittest.mock import Mock  # isort: skip
276

277
278
279
    _LIB = Mock(ctypes.CDLL)  # type: ignore
else:
    _LIB = _load_lib()
wxchan's avatar
wxchan committed
280

wxchan's avatar
wxchan committed
281

282
_NUMERIC_TYPES = (int, float, bool)
283
284


285
def _safe_call(ret: int) -> None:
286
287
    """Check the return value from C API call.

wxchan's avatar
wxchan committed
288
289
290
    Parameters
    ----------
    ret : int
291
        The return value from C API calls.
wxchan's avatar
wxchan committed
292
293
    """
    if ret != 0:
294
        raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
wxchan's avatar
wxchan committed
295

wxchan's avatar
wxchan committed
296

297
def _is_numeric(obj: Any) -> bool:
298
    """Check whether object is a number or not, include numpy number, etc."""
wxchan's avatar
wxchan committed
299
300
301
    try:
        float(obj)
        return True
wxchan's avatar
wxchan committed
302
303
304
    except (TypeError, ValueError):
        # TypeError: obj is not a string or a number
        # ValueError: invalid literal
wxchan's avatar
wxchan committed
305
306
        return False

wxchan's avatar
wxchan committed
307

308
def _is_numpy_1d_array(data: Any) -> bool:
309
    """Check whether data is a numpy 1-D array."""
310
    return isinstance(data, np.ndarray) and len(data.shape) == 1
wxchan's avatar
wxchan committed
311

wxchan's avatar
wxchan committed
312

313
def _is_numpy_column_array(data: Any) -> bool:
314
315
316
317
318
319
320
    """Check whether data is a column numpy array."""
    if not isinstance(data, np.ndarray):
        return False
    shape = data.shape
    return len(shape) == 2 and shape[1] == 1


321
def _cast_numpy_array_to_dtype(array: np.ndarray, dtype: "np.typing.DTypeLike") -> np.ndarray:
322
    """Cast numpy array to given dtype."""
323
324
325
326
327
    if array.dtype == dtype:
        return array
    return array.astype(dtype=dtype, copy=False)


328
def _is_1d_list(data: Any) -> bool:
329
    """Check whether data is a 1-D list."""
330
    return isinstance(data, list) and (not data or _is_numeric(data[0]))
wxchan's avatar
wxchan committed
331

wxchan's avatar
wxchan committed
332

333
def _is_list_of_numpy_arrays(data: Any) -> "TypeGuard[List[np.ndarray]]":
334
    return isinstance(data, list) and all(isinstance(x, np.ndarray) for x in data)
335
336
337


def _is_list_of_sequences(data: Any) -> "TypeGuard[List[Sequence]]":
338
    return isinstance(data, list) and all(isinstance(x, Sequence) for x in data)
339
340


341
342
def _is_1d_collection(data: Any) -> bool:
    """Check whether data is a 1-D collection."""
343
    return _is_numpy_1d_array(data) or _is_numpy_column_array(data) or _is_1d_list(data) or isinstance(data, pd_Series)
344
345


346
347
def _list_to_1d_numpy(
    data: Any,
348
    dtype: "np.typing.DTypeLike",
349
    name: str,
350
) -> np.ndarray:
351
    """Convert data to numpy 1-D array."""
352
    if _is_numpy_1d_array(data):
353
        return _cast_numpy_array_to_dtype(data, dtype)
354
    elif _is_numpy_column_array(data):
355
        _log_warning("Converting column-vector to 1d array")
356
        array = data.ravel()
357
        return _cast_numpy_array_to_dtype(array, dtype)
358
    elif _is_1d_list(data):
wxchan's avatar
wxchan committed
359
        return np.array(data, dtype=dtype, copy=False)
360
    elif isinstance(data, pd_Series):
361
        _check_for_bad_pandas_dtypes(data.to_frame().dtypes)
362
        return np.array(data, dtype=dtype, copy=False)  # SparseArray should be supported as well
wxchan's avatar
wxchan committed
363
    else:
364
365
366
        raise TypeError(
            f"Wrong type({type(data).__name__}) for {name}.\n" "It should be list, numpy 1-D array or pandas Series"
        )
wxchan's avatar
wxchan committed
367

wxchan's avatar
wxchan committed
368

369
370
371
372
373
374
375
def _is_numpy_2d_array(data: Any) -> bool:
    """Check whether data is a numpy 2-D array."""
    return isinstance(data, np.ndarray) and len(data.shape) == 2 and data.shape[1] > 1


def _is_2d_list(data: Any) -> bool:
    """Check whether data is a 2-D list."""
376
    return isinstance(data, list) and len(data) > 0 and _is_1d_list(data[0])
377
378
379
380


def _is_2d_collection(data: Any) -> bool:
    """Check whether data is a 2-D collection."""
381
    return _is_numpy_2d_array(data) or _is_2d_list(data) or isinstance(data, pd_DataFrame)
382
383


384
def _is_pyarrow_array(data: Any) -> "TypeGuard[Union[pa_Array, pa_ChunkedArray]]":
385
386
387
388
    """Check whether data is a PyArrow array."""
    return isinstance(data, (pa_Array, pa_ChunkedArray))


389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
def _is_pyarrow_table(data: Any) -> bool:
    """Check whether data is a PyArrow table."""
    return isinstance(data, pa_Table)


class _ArrowCArray:
    """Simple wrapper around the C representation of an Arrow type."""

    n_chunks: int
    chunks: arrow_cffi.CData
    schema: arrow_cffi.CData

    def __init__(self, n_chunks: int, chunks: arrow_cffi.CData, schema: arrow_cffi.CData):
        self.n_chunks = n_chunks
        self.chunks = chunks
        self.schema = schema

    @property
    def chunks_ptr(self) -> int:
        """Returns the address of the pointer to the list of chunks making up the array."""
        return int(arrow_cffi.cast("uintptr_t", arrow_cffi.addressof(self.chunks[0])))

    @property
    def schema_ptr(self) -> int:
        """Returns the address of the pointer to the schema of the array."""
        return int(arrow_cffi.cast("uintptr_t", self.schema))


def _export_arrow_to_c(data: pa_Table) -> _ArrowCArray:
    """Export an Arrow type to its C representation."""
    # Obtain objects to export
420
421
422
423
424
    if isinstance(data, pa_Array):
        export_objects = [data]
    elif isinstance(data, pa_ChunkedArray):
        export_objects = data.chunks
    elif isinstance(data, pa_Table):
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
        export_objects = data.to_batches()
    else:
        raise ValueError(f"data of type '{type(data)}' cannot be exported to Arrow")

    # Prepare export
    chunks = arrow_cffi.new("struct ArrowArray[]", len(export_objects))
    schema = arrow_cffi.new("struct ArrowSchema*")

    # Export all objects
    for i, obj in enumerate(export_objects):
        chunk_ptr = int(arrow_cffi.cast("uintptr_t", arrow_cffi.addressof(chunks[i])))
        if i == 0:
            schema_ptr = int(arrow_cffi.cast("uintptr_t", schema))
            obj._export_to_c(chunk_ptr, schema_ptr)
        else:
            obj._export_to_c(chunk_ptr)

    return _ArrowCArray(len(chunks), chunks, schema)


445
446
def _data_to_2d_numpy(
    data: Any,
447
    dtype: "np.typing.DTypeLike",
448
    name: str,
449
) -> np.ndarray:
450
451
    """Convert data to numpy 2-D array."""
    if _is_numpy_2d_array(data):
452
        return _cast_numpy_array_to_dtype(data, dtype)
453
454
455
    if _is_2d_list(data):
        return np.array(data, dtype=dtype)
    if isinstance(data, pd_DataFrame):
456
        _check_for_bad_pandas_dtypes(data.dtypes)
457
        return _cast_numpy_array_to_dtype(data.values, dtype)
458
459
460
461
    raise TypeError(
        f"Wrong type({type(data).__name__}) for {name}.\n"
        "It should be list of lists, numpy 2-D array or pandas DataFrame"
    )
462
463


464
def _cfloat32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
465
    """Convert a ctypes float pointer array to a numpy array."""
wxchan's avatar
wxchan committed
466
    if isinstance(cptr, ctypes.POINTER(ctypes.c_float)):
467
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
468
    else:
469
        raise RuntimeError("Expected float pointer")
wxchan's avatar
wxchan committed
470

Guolin Ke's avatar
Guolin Ke committed
471

472
def _cfloat64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
473
    """Convert a ctypes double pointer array to a numpy array."""
Guolin Ke's avatar
Guolin Ke committed
474
    if isinstance(cptr, ctypes.POINTER(ctypes.c_double)):
475
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
Guolin Ke's avatar
Guolin Ke committed
476
    else:
477
        raise RuntimeError("Expected double pointer")
Guolin Ke's avatar
Guolin Ke committed
478

wxchan's avatar
wxchan committed
479

480
def _cint32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
481
    """Convert a ctypes int pointer array to a numpy array."""
wxchan's avatar
wxchan committed
482
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int32)):
483
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
484
    else:
485
        raise RuntimeError("Expected int32 pointer")
486
487


488
def _cint64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
489
490
    """Convert a ctypes int pointer array to a numpy array."""
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int64)):
491
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
492
    else:
493
        raise RuntimeError("Expected int64 pointer")
wxchan's avatar
wxchan committed
494

wxchan's avatar
wxchan committed
495

496
def _c_str(string: str) -> ctypes.c_char_p:
497
    """Convert a Python string to C string."""
498
    return ctypes.c_char_p(string.encode("utf-8"))
wxchan's avatar
wxchan committed
499

wxchan's avatar
wxchan committed
500

501
def _c_array(ctype: type, values: List[Any]) -> ctypes.Array:
502
    """Convert a Python array to C array."""
503
    return (ctype * len(values))(*values)  # type: ignore[operator]
wxchan's avatar
wxchan committed
504

wxchan's avatar
wxchan committed
505

506
def _json_default_with_numpy(obj: Any) -> Any:
507
508
509
510
511
512
513
514
515
    """Convert numpy classes to JSON serializable objects."""
    if isinstance(obj, (np.integer, np.floating, np.bool_)):
        return obj.item()
    elif isinstance(obj, np.ndarray):
        return obj.tolist()
    else:
        return obj


516
517
518
519
520
521
522
523
def _to_string(x: Union[int, float, str, List]) -> str:
    if isinstance(x, list):
        val_list = ",".join(str(val) for val in x)
        return f"[{val_list}]"
    else:
        return str(x)


524
def _param_dict_to_str(data: Optional[Dict[str, Any]]) -> str:
525
    """Convert Python dictionary to string, which is passed to C API."""
526
    if data is None or not data:
wxchan's avatar
wxchan committed
527
528
529
        return ""
    pairs = []
    for key, val in data.items():
530
        if isinstance(val, (list, tuple, set)) or _is_numpy_1d_array(val):
531
            pairs.append(f"{key}={','.join(map(_to_string, val))}")
532
        elif isinstance(val, (str, Path, _NUMERIC_TYPES)) or _is_numeric(val):
533
            pairs.append(f"{key}={val}")
534
        elif val is not None:
535
536
            raise TypeError(f"Unknown type of parameter:{key}, got:{type(val).__name__}")
    return " ".join(pairs)
537

wxchan's avatar
wxchan committed
538

539
class _TempFile:
540
541
    """Proxy class to workaround errors on Windows."""

542
    def __enter__(self) -> "_TempFile":
543
544
        with NamedTemporaryFile(prefix="lightgbm_tmp_", delete=True) as f:
            self.name = f.name
545
            self.path = Path(self.name)
546
        return self
wxchan's avatar
wxchan committed
547

548
    def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None:
549
550
        if self.path.is_file():
            self.path.unlink()
551

wxchan's avatar
wxchan committed
552

553
class LightGBMError(Exception):
554
555
    """Error thrown by LightGBM."""

556
557
558
    pass


559
# DeprecationWarning is not shown by default, so let's create our own with higher level
560
561
# ref: https://peps.python.org/pep-0565/#additional-use-case-for-futurewarning
class LGBMDeprecationWarning(FutureWarning):
562
563
564
565
566
567
    """Custom deprecation warning."""

    pass


class _ConfigAliases:
568
569
570
571
    # lazy evaluation to allow import without dynamic library, e.g., for docs generation
    aliases = None

    @staticmethod
572
    def _get_all_param_aliases() -> Dict[str, List[str]]:
573
574
575
        buffer_len = 1 << 20
        tmp_out_len = ctypes.c_int64(0)
        string_buffer = ctypes.create_string_buffer(buffer_len)
576
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
577
578
579
580
581
582
583
        _safe_call(
            _LIB.LGBM_DumpParamAliases(
                ctypes.c_int64(buffer_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer,
            )
        )
584
585
586
587
        actual_len = tmp_out_len.value
        # if buffer length is not long enough, re-allocate a buffer
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
588
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
589
590
591
592
593
594
595
            _safe_call(
                _LIB.LGBM_DumpParamAliases(
                    ctypes.c_int64(actual_len),
                    ctypes.byref(tmp_out_len),
                    ptr_string_buffer,
                )
            )
596
        return json.loads(
597
            string_buffer.value.decode("utf-8"), object_hook=lambda obj: {k: [k] + v for k, v in obj.items()}
598
        )
599
600

    @classmethod
601
    def get(cls, *args: str) -> Set[str]:
602
603
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
604
605
        ret = set()
        for i in args:
606
            ret.update(cls.get_sorted(i))
607
608
        return ret

609
610
611
612
613
614
    @classmethod
    def get_sorted(cls, name: str) -> List[str]:
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
        return cls.aliases.get(name, [name])

615
    @classmethod
616
    def get_by_alias(cls, *args: str) -> Set[str]:
617
618
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
619
620
621
622
        ret = set(args)
        for arg in args:
            for aliases in cls.aliases.values():
                if arg in aliases:
623
                    ret.update(aliases)
624
625
626
                    break
        return ret

627

628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
def _choose_param_value(main_param_name: str, params: Dict[str, Any], default_value: Any) -> Dict[str, Any]:
    """Get a single parameter value, accounting for aliases.

    Parameters
    ----------
    main_param_name : str
        Name of the main parameter to get a value for. One of the keys of ``_ConfigAliases``.
    params : dict
        Dictionary of LightGBM parameters.
    default_value : Any
        Default value to use for the parameter, if none is found in ``params``.

    Returns
    -------
    params : dict
        A ``params`` dict with exactly one value for ``main_param_name``, and all aliases ``main_param_name`` removed.
        If both ``main_param_name`` and one or more aliases for it are found, the value of ``main_param_name`` will be preferred.
    """
    # avoid side effects on passed-in parameters
    params = deepcopy(params)

649
650
    aliases = _ConfigAliases.get_sorted(main_param_name)
    aliases = [a for a in aliases if a != main_param_name]
651
652

    # if main_param_name was provided, keep that value and remove all aliases
653
    if main_param_name in params.keys():
654
655
656
        for param in aliases:
            params.pop(param, None)
        return params
657

658
659
660
661
662
    # if main param name was not found, search for an alias
    for param in aliases:
        if param in params.keys():
            params[main_param_name] = params[param]
            break
663

664
665
666
667
668
669
670
    if main_param_name in params.keys():
        for param in aliases:
            params.pop(param, None)
        return params

    # neither of main_param_name, aliases were found
    params[main_param_name] = default_value
671
672
673
674

    return params


675
_MAX_INT32 = (1 << 31) - 1
676

677
"""Macro definition of data type in C API of LightGBM"""
678
679
680
681
_C_API_DTYPE_FLOAT32 = 0
_C_API_DTYPE_FLOAT64 = 1
_C_API_DTYPE_INT32 = 2
_C_API_DTYPE_INT64 = 3
Guolin Ke's avatar
Guolin Ke committed
682

683
"""Matrix is row major in Python"""
684
_C_API_IS_ROW_MAJOR = 1
wxchan's avatar
wxchan committed
685

686
"""Macro definition of prediction type in C API of LightGBM"""
687
688
689
690
_C_API_PREDICT_NORMAL = 0
_C_API_PREDICT_RAW_SCORE = 1
_C_API_PREDICT_LEAF_INDEX = 2
_C_API_PREDICT_CONTRIB = 3
wxchan's avatar
wxchan committed
691

692
"""Macro definition of sparse matrix type"""
693
694
_C_API_MATRIX_TYPE_CSR = 0
_C_API_MATRIX_TYPE_CSC = 1
695

696
"""Macro definition of feature importance type"""
697
698
_C_API_FEATURE_IMPORTANCE_SPLIT = 0
_C_API_FEATURE_IMPORTANCE_GAIN = 1
699

700
"""Data type of data field"""
701
702
703
704
_FIELD_TYPE_MAPPER = {
    "label": _C_API_DTYPE_FLOAT32,
    "weight": _C_API_DTYPE_FLOAT32,
    "init_score": _C_API_DTYPE_FLOAT64,
705
    "group": _C_API_DTYPE_INT32,
706
    "position": _C_API_DTYPE_INT32,
707
}
wxchan's avatar
wxchan committed
708

709
"""String name to int feature importance type mapper"""
710
711
_FEATURE_IMPORTANCE_TYPE_MAPPER = {
    "split": _C_API_FEATURE_IMPORTANCE_SPLIT,
712
    "gain": _C_API_FEATURE_IMPORTANCE_GAIN,
713
}
714

wxchan's avatar
wxchan committed
715

716
def _convert_from_sliced_object(data: np.ndarray) -> np.ndarray:
717
    """Fix the memory of multi-dimensional sliced object."""
718
    if isinstance(data, np.ndarray) and isinstance(data.base, np.ndarray):
719
        if not data.flags.c_contiguous:
720
721
722
723
            _log_warning(
                "Usage of np.ndarray subset (sliced data) is not recommended "
                "due to it will double the peak memory cost in LightGBM."
            )
724
725
726
727
            return np.copy(data)
    return data


728
def _c_float_array(data: np.ndarray) -> Tuple[_ctypes_float_ptr, int, np.ndarray]:
729
    """Get pointer of float numpy array / list."""
730
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
731
        data = np.array(data, copy=False)
732
    if _is_numpy_1d_array(data):
733
        data = _convert_from_sliced_object(data)
734
        assert data.flags.c_contiguous
735
        ptr_data: _ctypes_float_ptr
wxchan's avatar
wxchan committed
736
737
        if data.dtype == np.float32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
738
            type_data = _C_API_DTYPE_FLOAT32
wxchan's avatar
wxchan committed
739
740
        elif data.dtype == np.float64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_double))
741
            type_data = _C_API_DTYPE_FLOAT64
wxchan's avatar
wxchan committed
742
        else:
743
            raise TypeError(f"Expected np.float32 or np.float64, met type({data.dtype})")
wxchan's avatar
wxchan committed
744
    else:
745
        raise TypeError(f"Unknown type({type(data).__name__})")
746
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
747

wxchan's avatar
wxchan committed
748

749
def _c_int_array(data: np.ndarray) -> Tuple[_ctypes_int_ptr, int, np.ndarray]:
750
    """Get pointer of int numpy array / list."""
751
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
752
        data = np.array(data, copy=False)
753
    if _is_numpy_1d_array(data):
754
        data = _convert_from_sliced_object(data)
755
        assert data.flags.c_contiguous
756
        ptr_data: _ctypes_int_ptr
wxchan's avatar
wxchan committed
757
758
        if data.dtype == np.int32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int32))
759
            type_data = _C_API_DTYPE_INT32
wxchan's avatar
wxchan committed
760
761
        elif data.dtype == np.int64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int64))
762
            type_data = _C_API_DTYPE_INT64
wxchan's avatar
wxchan committed
763
        else:
764
            raise TypeError(f"Expected np.int32 or np.int64, met type({data.dtype})")
wxchan's avatar
wxchan committed
765
    else:
766
        raise TypeError(f"Unknown type({type(data).__name__})")
767
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
768

wxchan's avatar
wxchan committed
769

770
def _is_allowed_numpy_dtype(dtype: type) -> bool:
771
772
    float128 = getattr(np, "float128", type(None))
    return issubclass(dtype, (np.integer, np.floating, np.bool_)) and not issubclass(dtype, (np.timedelta64, float128))
773
774


775
def _check_for_bad_pandas_dtypes(pandas_dtypes_series: pd_Series) -> None:
776
    bad_pandas_dtypes = [
777
        f"{column_name}: {pandas_dtype}"
778
        for column_name, pandas_dtype in pandas_dtypes_series.items()
779
        if not _is_allowed_numpy_dtype(pandas_dtype.type)
780
781
    ]
    if bad_pandas_dtypes:
782
783
784
785
        raise ValueError(
            'pandas dtypes must be int, float or bool.\n'
            f'Fields with bad pandas dtypes: {", ".join(bad_pandas_dtypes)}'
        )
786
787


788
789
def _pandas_to_numpy(
    data: pd_DataFrame,
790
    target_dtype: "np.typing.DTypeLike",
791
792
793
794
795
796
797
798
799
800
801
802
803
804
) -> np.ndarray:
    _check_for_bad_pandas_dtypes(data.dtypes)
    try:
        # most common case (no nullable dtypes)
        return data.to_numpy(dtype=target_dtype, copy=False)
    except TypeError:
        # 1.0 <= pd version < 1.1 and nullable dtypes, least common case
        # raises error because array is casted to type(pd.NA) and there's no na_value argument
        return data.astype(target_dtype, copy=False).values
    except ValueError:
        # data has nullable dtypes, but we can specify na_value argument and copy will be made
        return data.to_numpy(dtype=target_dtype, na_value=np.nan)


805
def _data_from_pandas(
806
807
808
    data: pd_DataFrame,
    feature_name: _LGBM_FeatureNameConfiguration,
    categorical_feature: _LGBM_CategoricalFeatureConfiguration,
809
    pandas_categorical: Optional[List[List]],
810
) -> Tuple[np.ndarray, List[str], Union[List[str], List[int]], List[List]]:
811
    if len(data.shape) != 2 or data.shape[0] < 1:
812
        raise ValueError("Input data must be 2 dimensional and non empty.")
813

814
815
816
817
    # take shallow copy in case we modify categorical columns
    # whole column modifications don't change the original df
    data = data.copy(deep=False)

818
    # determine feature names
819
    if feature_name == "auto":
820
821
822
823
        feature_name = [str(col) for col in data.columns]

    # determine categorical features
    cat_cols = [col for col, dtype in zip(data.columns, data.dtypes) if isinstance(dtype, pd_CategoricalDtype)]
824
    cat_cols_not_ordered: List[str] = [col for col in cat_cols if not data[col].cat.ordered]
825
826
    if pandas_categorical is None:  # train dataset
        pandas_categorical = [list(data[col].cat.categories) for col in cat_cols]
827
    else:
828
        if len(cat_cols) != len(pandas_categorical):
829
            raise ValueError("train and valid dataset categorical_feature do not match.")
830
831
832
833
834
        for col, category in zip(cat_cols, pandas_categorical):
            if list(data[col].cat.categories) != list(category):
                data[col] = data[col].cat.set_categories(category)
    if len(cat_cols):  # cat_cols is list
        data[cat_cols] = data[cat_cols].apply(lambda x: x.cat.codes).replace({-1: np.nan})
835
836

    # use cat cols from DataFrame
837
    if categorical_feature == "auto":
838
839
840
        categorical_feature = cat_cols_not_ordered

    df_dtypes = [dtype.type for dtype in data.dtypes]
841
842
    # so that the target dtype considers floats
    df_dtypes.append(np.float32)
843
    target_dtype = np.result_type(*df_dtypes)
844
845
846
847
848

    return (
        _pandas_to_numpy(data, target_dtype=target_dtype),
        feature_name,
        categorical_feature,
849
        pandas_categorical,
850
    )
851
852


853
854
def _dump_pandas_categorical(
    pandas_categorical: Optional[List[List]],
855
    file_name: Optional[Union[str, Path]] = None,
856
) -> str:
857
    categorical_json = json.dumps(pandas_categorical, default=_json_default_with_numpy)
858
    pandas_str = f"\npandas_categorical:{categorical_json}\n"
859
    if file_name is not None:
860
        with open(file_name, "a") as f:
861
862
863
864
            f.write(pandas_str)
    return pandas_str


865
866
def _load_pandas_categorical(
    file_name: Optional[Union[str, Path]] = None,
867
    model_str: Optional[str] = None,
868
) -> Optional[List[List]]:
869
    pandas_key = "pandas_categorical:"
870
    offset = -len(pandas_key)
871
    if file_name is not None:
872
        max_offset = -getsize(file_name)
873
        with open(file_name, "rb") as f:
874
875
876
            while True:
                if offset < max_offset:
                    offset = max_offset
877
                f.seek(offset, SEEK_END)
878
879
880
881
                lines = f.readlines()
                if len(lines) >= 2:
                    break
                offset *= 2
882
        last_line = lines[-1].decode("utf-8").strip()
883
        if not last_line.startswith(pandas_key):
884
            last_line = lines[-2].decode("utf-8").strip()
885
    elif model_str is not None:
886
        idx = model_str.rfind("\n", 0, offset)
887
888
        last_line = model_str[idx:].strip()
    if last_line.startswith(pandas_key):
889
        return json.loads(last_line[len(pandas_key) :])
890
891
    else:
        return None
892
893


894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
class Sequence(abc.ABC):
    """
    Generic data access interface.

    Object should support the following operations:

    .. code-block::

        # Get total row number.
        >>> len(seq)
        # Random access by row index. Used for data sampling.
        >>> seq[10]
        # Range data access. Used to read data in batch when constructing Dataset.
        >>> seq[0:100]
        # Optionally specify batch_size to control range data read size.
        >>> seq.batch_size

    - With random access, **data sampling does not need to go through all data**.
    - With range data access, there's **no need to read all data into memory thus reduce memory usage**.

914
915
    .. versionadded:: 3.3.0

916
917
918
919
920
921
922
923
924
    Attributes
    ----------
    batch_size : int
        Default size of a batch.
    """

    batch_size = 4096  # Defaults to read 4K rows in each batch.

    @abc.abstractmethod
925
    def __getitem__(self, idx: Union[int, slice, List[int]]) -> np.ndarray:
926
927
928
929
930
931
932
        """Return data for given row index.

        A basic implementation should look like this:

        .. code-block:: python

            if isinstance(idx, numbers.Integral):
933
                return self._get_one_line(idx)
934
            elif isinstance(idx, slice):
935
936
                return np.stack([self._get_one_line(i) for i in range(idx.start, idx.stop)])
            elif isinstance(idx, list):
937
                # Only required if using ``Dataset.subset()``.
938
                return np.array([self._get_one_line(i) for i in idx])
939
            else:
940
                raise TypeError(f"Sequence index must be integer, slice or list, got {type(idx).__name__}")
941
942
943

        Parameters
        ----------
944
        idx : int, slice[int], list[int]
945
946
947
948
            Item index.

        Returns
        -------
949
        result : numpy 1-D array or numpy 2-D array
950
            1-D array if idx is int, 2-D array if idx is slice or list.
951
952
953
954
955
956
957
958
959
        """
        raise NotImplementedError("Sub-classes of lightgbm.Sequence must implement __getitem__()")

    @abc.abstractmethod
    def __len__(self) -> int:
        """Return row count of this sequence."""
        raise NotImplementedError("Sub-classes of lightgbm.Sequence must implement __len__()")


960
class _InnerPredictor:
961
962
963
964
965
    """_InnerPredictor of LightGBM.

    Not exposed to user.
    Used only for prediction, usually used for continued training.

Nikita Titov's avatar
Nikita Titov committed
966
967
968
    .. note::

        Can be converted from Booster, but cannot be converted to Booster.
Guolin Ke's avatar
Guolin Ke committed
969
    """
970

971
972
    def __init__(
        self,
973
974
975
        booster_handle: _BoosterHandle,
        pandas_categorical: Optional[List[List]],
        pred_parameter: Dict[str, Any],
976
        manage_handle: bool,
977
    ):
978
        """Initialize the _InnerPredictor.
wxchan's avatar
wxchan committed
979
980
981

        Parameters
        ----------
982
        booster_handle : object
983
            Handle of Booster.
984
985
986
987
        pandas_categorical : list of list, or None
            If provided, list of categories for ``pandas`` categorical columns.
            Where the ``i``th element of the list contains the categories for the ``i``th categorical feature.
        pred_parameter : dict
988
            Other parameters for the prediction.
989
990
        manage_handle : bool
            If ``True``, free the corresponding Booster on the C++ side when this Python object is deleted.
wxchan's avatar
wxchan committed
991
        """
992
993
994
995
996
997
998
999
        self._handle = booster_handle
        self.__is_manage_handle = manage_handle
        self.pandas_categorical = pandas_categorical
        self.pred_parameter = _param_dict_to_str(pred_parameter)

        out_num_class = ctypes.c_int(0)
        _safe_call(
            _LIB.LGBM_BoosterGetNumClasses(
1000
                self._handle,
1001
                ctypes.byref(out_num_class),
1002
1003
1004
            )
        )
        self.num_class = out_num_class.value
wxchan's avatar
wxchan committed
1005

1006
1007
1008
1009
    @classmethod
    def from_booster(
        cls,
        booster: "Booster",
1010
        pred_parameter: Dict[str, Any],
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
    ) -> "_InnerPredictor":
        """Initialize an ``_InnerPredictor`` from a ``Booster``.

        Parameters
        ----------
        booster : Booster
            Booster.
        pred_parameter : dict
            Other parameters for the prediction.
        """
        out_cur_iter = ctypes.c_int(0)
        _safe_call(
            _LIB.LGBM_BoosterGetCurrentIteration(
                booster._handle,
1025
                ctypes.byref(out_cur_iter),
1026
1027
1028
1029
1030
1031
            )
        )
        return cls(
            booster_handle=booster._handle,
            pandas_categorical=booster.pandas_categorical,
            pred_parameter=pred_parameter,
1032
            manage_handle=False,
1033
1034
1035
1036
1037
1038
        )

    @classmethod
    def from_model_file(
        cls,
        model_file: Union[str, Path],
1039
        pred_parameter: Dict[str, Any],
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
    ) -> "_InnerPredictor":
        """Initialize an ``_InnerPredictor`` from a text file containing a LightGBM model.

        Parameters
        ----------
        model_file : str or pathlib.Path
            Path to the model file.
        pred_parameter : dict
            Other parameters for the prediction.
        """
        booster_handle = ctypes.c_void_p()
        out_num_iterations = ctypes.c_int(0)
        _safe_call(
            _LIB.LGBM_BoosterCreateFromModelfile(
                _c_str(str(model_file)),
                ctypes.byref(out_num_iterations),
1056
                ctypes.byref(booster_handle),
1057
1058
1059
1060
1061
1062
            )
        )
        return cls(
            booster_handle=booster_handle,
            pandas_categorical=_load_pandas_categorical(file_name=model_file),
            pred_parameter=pred_parameter,
1063
            manage_handle=True,
1064
        )
cbecker's avatar
cbecker committed
1065

1066
    def __del__(self) -> None:
1067
1068
        try:
            if self.__is_manage_handle:
1069
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
1070
1071
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
1072

1073
    def __getstate__(self) -> Dict[str, Any]:
1074
        this = self.__dict__.copy()
1075
1076
        this.pop("handle", None)
        this.pop("_handle", None)
1077
1078
        return this

1079
1080
    def predict(
        self,
1081
        data: _LGBM_PredictDataType,
1082
1083
1084
1085
1086
1087
        start_iteration: int = 0,
        num_iteration: int = -1,
        raw_score: bool = False,
        pred_leaf: bool = False,
        pred_contrib: bool = False,
        data_has_header: bool = False,
1088
        validate_features: bool = False,
1089
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
1090
        """Predict logic.
wxchan's avatar
wxchan committed
1091
1092
1093

        Parameters
        ----------
1094
        data : str, pathlib.Path, numpy array, pandas DataFrame, pyarrow Table, H2O DataTable's Frame or scipy.sparse
1095
            Data source for prediction.
1096
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
1097
1098
        start_iteration : int, optional (default=0)
            Start index of the iteration to predict.
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
        num_iteration : int, optional (default=-1)
            Iteration used for prediction.
        raw_score : bool, optional (default=False)
            Whether to predict raw scores.
        pred_leaf : bool, optional (default=False)
            Whether to predict leaf index.
        pred_contrib : bool, optional (default=False)
            Whether to predict feature contributions.
        data_has_header : bool, optional (default=False)
            Whether data has header.
            Used only for txt data.
1110
1111
1112
        validate_features : bool, optional (default=False)
            If True, ensure that the features used to predict match the ones used to train.
            Used only if data is pandas DataFrame.
wxchan's avatar
wxchan committed
1113

1114
1115
            .. versionadded:: 4.0.0

wxchan's avatar
wxchan committed
1116
1117
        Returns
        -------
1118
        result : numpy array, scipy.sparse or list of scipy.sparse
1119
            Prediction result.
1120
            Can be sparse or a list of sparse objects (each element represents predictions for one class) for feature contributions (when ``pred_contrib=True``).
wxchan's avatar
wxchan committed
1121
        """
wxchan's avatar
wxchan committed
1122
        if isinstance(data, Dataset):
1123
            raise TypeError("Cannot use Dataset instance for prediction, please use raw data instead")
1124
1125
1126
        elif isinstance(data, pd_DataFrame) and validate_features:
            data_names = [str(x) for x in data.columns]
            ptr_names = (ctypes.c_char_p * len(data_names))()
1127
            ptr_names[:] = [x.encode("utf-8") for x in data_names]
1128
1129
            _safe_call(
                _LIB.LGBM_BoosterValidateFeatureNames(
1130
                    self._handle,
1131
1132
1133
1134
                    ptr_names,
                    ctypes.c_int(len(data_names)),
                )
            )
1135
1136
1137
1138
1139
1140

        if isinstance(data, pd_DataFrame):
            data = _data_from_pandas(
                data=data,
                feature_name="auto",
                categorical_feature="auto",
1141
                pandas_categorical=self.pandas_categorical,
1142
1143
            )[0]

1144
        predict_type = _C_API_PREDICT_NORMAL
wxchan's avatar
wxchan committed
1145
        if raw_score:
1146
            predict_type = _C_API_PREDICT_RAW_SCORE
wxchan's avatar
wxchan committed
1147
        if pred_leaf:
1148
            predict_type = _C_API_PREDICT_LEAF_INDEX
1149
        if pred_contrib:
1150
            predict_type = _C_API_PREDICT_CONTRIB
cbecker's avatar
cbecker committed
1151

1152
        if isinstance(data, (str, Path)):
1153
            with _TempFile() as f:
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
                _safe_call(
                    _LIB.LGBM_BoosterPredictForFile(
                        self._handle,
                        _c_str(str(data)),
                        ctypes.c_int(data_has_header),
                        ctypes.c_int(predict_type),
                        ctypes.c_int(start_iteration),
                        ctypes.c_int(num_iteration),
                        _c_str(self.pred_parameter),
                        _c_str(f.name),
                    )
                )
1166
1167
                preds = np.loadtxt(f.name, dtype=np.float64)
                nrow = preds.shape[0]
wxchan's avatar
wxchan committed
1168
        elif isinstance(data, scipy.sparse.csr_matrix):
1169
1170
1171
1172
            preds, nrow = self.__pred_for_csr(
                csr=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1173
                predict_type=predict_type,
1174
            )
Guolin Ke's avatar
Guolin Ke committed
1175
        elif isinstance(data, scipy.sparse.csc_matrix):
1176
1177
1178
1179
            preds, nrow = self.__pred_for_csc(
                csc=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1180
                predict_type=predict_type,
1181
            )
wxchan's avatar
wxchan committed
1182
        elif isinstance(data, np.ndarray):
1183
1184
1185
1186
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1187
                predict_type=predict_type,
1188
            )
1189
1190
1191
1192
1193
        elif _is_pyarrow_table(data):
            preds, nrow = self.__pred_for_pyarrow_table(
                table=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1194
                predict_type=predict_type,
1195
            )
1196
1197
1198
        elif isinstance(data, list):
            try:
                data = np.array(data)
1199
            except BaseException as err:
1200
                raise ValueError("Cannot convert data list to numpy array.") from err
1201
1202
1203
1204
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1205
                predict_type=predict_type,
1206
            )
1207
        elif isinstance(data, dt_DataTable):
1208
1209
1210
1211
            preds, nrow = self.__pred_for_np2d(
                mat=data.to_numpy(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1212
                predict_type=predict_type,
1213
            )
wxchan's avatar
wxchan committed
1214
1215
        else:
            try:
1216
                _log_warning("Converting data to scipy sparse matrix.")
wxchan's avatar
wxchan committed
1217
                csr = scipy.sparse.csr_matrix(data)
1218
            except BaseException as err:
1219
                raise TypeError(f"Cannot predict data for type {type(data).__name__}") from err
1220
1221
1222
1223
            preds, nrow = self.__pred_for_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1224
                predict_type=predict_type,
1225
            )
wxchan's avatar
wxchan committed
1226
1227
        if pred_leaf:
            preds = preds.astype(np.int32)
1228
        is_sparse = isinstance(preds, (list, scipy.sparse.spmatrix))
1229
        if not is_sparse and preds.size != nrow:
wxchan's avatar
wxchan committed
1230
            if preds.size % nrow == 0:
1231
                preds = preds.reshape(nrow, -1)
wxchan's avatar
wxchan committed
1232
            else:
1233
                raise ValueError(f"Length of predict result ({preds.size}) cannot be divide nrow ({nrow})")
wxchan's avatar
wxchan committed
1234
1235
        return preds

1236
1237
1238
1239
1240
    def __get_num_preds(
        self,
        start_iteration: int,
        num_iteration: int,
        nrow: int,
1241
        predict_type: int,
1242
    ) -> int:
1243
        """Get size of prediction result."""
1244
        if nrow > _MAX_INT32:
1245
1246
1247
1248
1249
1250
            raise LightGBMError(
                "LightGBM cannot perform prediction for data "
                f"with number of rows greater than MAX_INT32 ({_MAX_INT32}).\n"
                "You can split your data into chunks "
                "and then concatenate predictions for them"
            )
Guolin Ke's avatar
Guolin Ke committed
1251
        n_preds = ctypes.c_int64(0)
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
        _safe_call(
            _LIB.LGBM_BoosterCalcNumPredict(
                self._handle,
                ctypes.c_int(nrow),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                ctypes.byref(n_preds),
            )
        )
Guolin Ke's avatar
Guolin Ke committed
1262
        return n_preds.value
wxchan's avatar
wxchan committed
1263

1264
1265
1266
1267
1268
1269
    def __inner_predict_np2d(
        self,
        mat: np.ndarray,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
1270
        preds: Optional[np.ndarray],
1271
1272
1273
1274
1275
1276
    ) -> Tuple[np.ndarray, int]:
        if mat.dtype == np.float32 or mat.dtype == np.float64:
            data = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
        else:  # change non-float data to float data, need to copy
            data = np.array(mat.reshape(mat.size), dtype=np.float32)
        ptr_data, type_ptr_data, _ = _c_float_array(data)
1277
1278
1279
1280
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=mat.shape[0],
1281
            predict_type=predict_type,
1282
        )
1283
1284
1285
1286
1287
        if preds is None:
            preds = np.empty(n_preds, dtype=np.float64)
        elif len(preds.shape) != 1 or len(preds) != n_preds:
            raise ValueError("Wrong length of pre-allocated predict array")
        out_num_preds = ctypes.c_int64(0)
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
        _safe_call(
            _LIB.LGBM_BoosterPredictForMat(
                self._handle,
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int32(mat.shape[0]),
                ctypes.c_int32(mat.shape[1]),
                ctypes.c_int(_C_API_IS_ROW_MAJOR),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.byref(out_num_preds),
                preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
            )
        )
1304
1305
1306
1307
1308
1309
1310
1311
1312
        if n_preds != out_num_preds.value:
            raise ValueError("Wrong length for predict results")
        return preds, mat.shape[0]

    def __pred_for_np2d(
        self,
        mat: np.ndarray,
        start_iteration: int,
        num_iteration: int,
1313
        predict_type: int,
1314
    ) -> Tuple[np.ndarray, int]:
1315
        """Predict for a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
1316
        if len(mat.shape) != 2:
1317
            raise ValueError("Input numpy.ndarray or list must be 2 dimensional")
wxchan's avatar
wxchan committed
1318

1319
        nrow = mat.shape[0]
1320
1321
        if nrow > _MAX_INT32:
            sections = np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)
1322
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1323
1324
1325
1326
            n_preds = [
                self.__get_num_preds(start_iteration, num_iteration, i, predict_type)
                for i in np.diff([0] + list(sections) + [nrow])
            ]
1327
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1328
            preds = np.empty(sum(n_preds), dtype=np.float64)
1329
1330
1331
            for chunk, (start_idx_pred, end_idx_pred) in zip(
                np.array_split(mat, sections), zip(n_preds_sections, n_preds_sections[1:])
            ):
1332
                # avoid memory consumption by arrays concatenation operations
1333
1334
1335
1336
1337
                self.__inner_predict_np2d(
                    mat=chunk,
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
1338
                    preds=preds[start_idx_pred:end_idx_pred],
1339
                )
1340
            return preds, nrow
wxchan's avatar
wxchan committed
1341
        else:
1342
1343
1344
1345
1346
            return self.__inner_predict_np2d(
                mat=mat,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
1347
                preds=None,
1348
            )
wxchan's avatar
wxchan committed
1349

1350
1351
1352
    def __create_sparse_native(
        self,
        cs: Union[scipy.sparse.csc_matrix, scipy.sparse.csr_matrix],
1353
1354
1355
1356
1357
1358
        out_shape: np.ndarray,
        out_ptr_indptr: "ctypes._Pointer",
        out_ptr_indices: "ctypes._Pointer",
        out_ptr_data: "ctypes._Pointer",
        indptr_type: int,
        data_type: int,
1359
        is_csr: bool,
1360
    ) -> Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]]:
1361
1362
1363
        # create numpy array from output arrays
        data_indices_len = out_shape[0]
        indptr_len = out_shape[1]
1364
        if indptr_type == _C_API_DTYPE_INT32:
1365
            out_indptr = _cint32_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1366
        elif indptr_type == _C_API_DTYPE_INT64:
1367
            out_indptr = _cint64_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1368
1369
        else:
            raise TypeError("Expected int32 or int64 type for indptr")
1370
        if data_type == _C_API_DTYPE_FLOAT32:
1371
            out_data = _cfloat32_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1372
        elif data_type == _C_API_DTYPE_FLOAT64:
1373
            out_data = _cfloat64_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1374
1375
        else:
            raise TypeError("Expected float32 or float64 type for data")
1376
        out_indices = _cint32_array_to_numpy(cptr=out_ptr_indices, length=data_indices_len)
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
        # break up indptr based on number of rows (note more than one matrix in multiclass case)
        per_class_indptr_shape = cs.indptr.shape[0]
        # for CSC there is extra column added
        if not is_csr:
            per_class_indptr_shape += 1
        out_indptr_arrays = np.split(out_indptr, out_indptr.shape[0] / per_class_indptr_shape)
        # reformat output into a csr or csc matrix or list of csr or csc matrices
        cs_output_matrices = []
        offset = 0
        for cs_indptr in out_indptr_arrays:
            matrix_indptr_len = cs_indptr[cs_indptr.shape[0] - 1]
1388
1389
            cs_indices = out_indices[offset + cs_indptr[0] : offset + matrix_indptr_len]
            cs_data = out_data[offset + cs_indptr[0] : offset + matrix_indptr_len]
1390
1391
1392
1393
1394
1395
1396
1397
1398
            offset += matrix_indptr_len
            # same shape as input csr or csc matrix except extra column for expected value
            cs_shape = [cs.shape[0], cs.shape[1] + 1]
            # note: make sure we copy data as it will be deallocated next
            if is_csr:
                cs_output_matrices.append(scipy.sparse.csr_matrix((cs_data, cs_indices, cs_indptr), cs_shape))
            else:
                cs_output_matrices.append(scipy.sparse.csc_matrix((cs_data, cs_indices, cs_indptr), cs_shape))
        # free the temporary native indptr, indices, and data
1399
1400
1401
1402
1403
1404
1405
1406
1407
        _safe_call(
            _LIB.LGBM_BoosterFreePredictSparse(
                out_ptr_indptr,
                out_ptr_indices,
                out_ptr_data,
                ctypes.c_int(indptr_type),
                ctypes.c_int(data_type),
            )
        )
1408
1409
1410
1411
        if len(cs_output_matrices) == 1:
            return cs_output_matrices[0]
        return cs_output_matrices

1412
1413
1414
1415
1416
1417
    def __inner_predict_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
1418
        preds: Optional[np.ndarray],
1419
1420
    ) -> Tuple[np.ndarray, int]:
        nrow = len(csr.indptr) - 1
1421
1422
1423
1424
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
1425
            predict_type=predict_type,
1426
        )
1427
1428
1429
1430
1431
        if preds is None:
            preds = np.empty(n_preds, dtype=np.float64)
        elif len(preds.shape) != 1 or len(preds) != n_preds:
            raise ValueError("Wrong length of pre-allocated predict array")
        out_num_preds = ctypes.c_int64(0)
wxchan's avatar
wxchan committed
1432

1433
1434
        ptr_indptr, type_ptr_indptr, _ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
1435

1436
1437
1438
        assert csr.shape[1] <= _MAX_INT32
        csr_indices = csr.indices.astype(np.int32, copy=False)

1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
        _safe_call(
            _LIB.LGBM_BoosterPredictForCSR(
                self._handle,
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csr.indptr)),
                ctypes.c_int64(len(csr.data)),
                ctypes.c_int64(csr.shape[1]),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.byref(out_num_preds),
                preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
            )
        )
1458
1459
1460
1461
1462
1463
1464
1465
1466
        if n_preds != out_num_preds.value:
            raise ValueError("Wrong length for predict results")
        return preds, nrow

    def __inner_predict_csr_sparse(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
1467
        predict_type: int,
1468
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1469
1470
1471
1472
        ptr_indptr, type_ptr_indptr, __ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
        csr_indices = csr.indices.astype(np.int32, copy=False)
        matrix_type = _C_API_MATRIX_TYPE_CSR
1473
        out_ptr_indptr: _ctypes_int_ptr
1474
1475
1476
1477
1478
        if type_ptr_indptr == _C_API_DTYPE_INT32:
            out_ptr_indptr = ctypes.POINTER(ctypes.c_int32)()
        else:
            out_ptr_indptr = ctypes.POINTER(ctypes.c_int64)()
        out_ptr_indices = ctypes.POINTER(ctypes.c_int32)()
1479
        out_ptr_data: _ctypes_float_ptr
1480
1481
1482
1483
1484
        if type_ptr_data == _C_API_DTYPE_FLOAT32:
            out_ptr_data = ctypes.POINTER(ctypes.c_float)()
        else:
            out_ptr_data = ctypes.POINTER(ctypes.c_double)()
        out_shape = np.empty(2, dtype=np.int64)
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
        _safe_call(
            _LIB.LGBM_BoosterPredictSparseOutput(
                self._handle,
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csr.indptr)),
                ctypes.c_int64(len(csr.data)),
                ctypes.c_int64(csr.shape[1]),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.c_int(matrix_type),
                out_shape.ctypes.data_as(ctypes.POINTER(ctypes.c_int64)),
                ctypes.byref(out_ptr_indptr),
                ctypes.byref(out_ptr_indices),
                ctypes.byref(out_ptr_data),
            )
        )
1507
1508
1509
1510
1511
1512
1513
1514
        matrices = self.__create_sparse_native(
            cs=csr,
            out_shape=out_shape,
            out_ptr_indptr=out_ptr_indptr,
            out_ptr_indices=out_ptr_indices,
            out_ptr_data=out_ptr_data,
            indptr_type=type_ptr_indptr,
            data_type=type_ptr_data,
1515
            is_csr=True,
1516
1517
1518
1519
        )
        nrow = len(csr.indptr) - 1
        return matrices, nrow

1520
1521
1522
1523
1524
    def __pred_for_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
1525
        predict_type: int,
1526
    ) -> Tuple[np.ndarray, int]:
1527
        """Predict for a CSR data."""
1528
        if predict_type == _C_API_PREDICT_CONTRIB:
1529
1530
1531
1532
            return self.__inner_predict_csr_sparse(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1533
                predict_type=predict_type,
1534
            )
1535
        nrow = len(csr.indptr) - 1
1536
1537
        if nrow > _MAX_INT32:
            sections = [0] + list(np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)) + [nrow]
1538
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1539
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff(sections)]
1540
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1541
            preds = np.empty(sum(n_preds), dtype=np.float64)
1542
1543
1544
            for (start_idx, end_idx), (start_idx_pred, end_idx_pred) in zip(
                zip(sections, sections[1:]), zip(n_preds_sections, n_preds_sections[1:])
            ):
1545
                # avoid memory consumption by arrays concatenation operations
1546
1547
1548
1549
1550
                self.__inner_predict_csr(
                    csr=csr[start_idx:end_idx],
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
1551
                    preds=preds[start_idx_pred:end_idx_pred],
1552
                )
1553
1554
            return preds, nrow
        else:
1555
1556
1557
1558
1559
            return self.__inner_predict_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
1560
                preds=None,
1561
1562
1563
1564
            )

    def __inner_predict_sparse_csc(
        self,
1565
1566
1567
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
1568
        predict_type: int,
1569
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1570
1571
1572
1573
        ptr_indptr, type_ptr_indptr, __ = _c_int_array(csc.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csc.data)
        csc_indices = csc.indices.astype(np.int32, copy=False)
        matrix_type = _C_API_MATRIX_TYPE_CSC
1574
        out_ptr_indptr: _ctypes_int_ptr
1575
1576
1577
1578
1579
        if type_ptr_indptr == _C_API_DTYPE_INT32:
            out_ptr_indptr = ctypes.POINTER(ctypes.c_int32)()
        else:
            out_ptr_indptr = ctypes.POINTER(ctypes.c_int64)()
        out_ptr_indices = ctypes.POINTER(ctypes.c_int32)()
1580
        out_ptr_data: _ctypes_float_ptr
1581
1582
1583
1584
1585
        if type_ptr_data == _C_API_DTYPE_FLOAT32:
            out_ptr_data = ctypes.POINTER(ctypes.c_float)()
        else:
            out_ptr_data = ctypes.POINTER(ctypes.c_double)()
        out_shape = np.empty(2, dtype=np.int64)
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
        _safe_call(
            _LIB.LGBM_BoosterPredictSparseOutput(
                self._handle,
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csc.indptr)),
                ctypes.c_int64(len(csc.data)),
                ctypes.c_int64(csc.shape[0]),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.c_int(matrix_type),
                out_shape.ctypes.data_as(ctypes.POINTER(ctypes.c_int64)),
                ctypes.byref(out_ptr_indptr),
                ctypes.byref(out_ptr_indices),
                ctypes.byref(out_ptr_data),
            )
        )
1608
1609
1610
1611
1612
1613
1614
1615
        matrices = self.__create_sparse_native(
            cs=csc,
            out_shape=out_shape,
            out_ptr_indptr=out_ptr_indptr,
            out_ptr_indices=out_ptr_indices,
            out_ptr_data=out_ptr_data,
            indptr_type=type_ptr_indptr,
            data_type=type_ptr_data,
1616
            is_csr=False,
1617
1618
1619
        )
        nrow = csc.shape[0]
        return matrices, nrow
Guolin Ke's avatar
Guolin Ke committed
1620

1621
1622
1623
1624
1625
    def __pred_for_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
1626
        predict_type: int,
1627
    ) -> Tuple[np.ndarray, int]:
1628
        """Predict for a CSC data."""
Guolin Ke's avatar
Guolin Ke committed
1629
        nrow = csc.shape[0]
1630
        if nrow > _MAX_INT32:
1631
1632
1633
1634
            return self.__pred_for_csr(
                csr=csc.tocsr(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1635
                predict_type=predict_type,
1636
            )
1637
        if predict_type == _C_API_PREDICT_CONTRIB:
1638
1639
1640
1641
            return self.__inner_predict_sparse_csc(
                csc=csc,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1642
                predict_type=predict_type,
1643
            )
1644
1645
1646
1647
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
1648
            predict_type=predict_type,
1649
        )
1650
        preds = np.empty(n_preds, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
1651
1652
        out_num_preds = ctypes.c_int64(0)

1653
1654
        ptr_indptr, type_ptr_indptr, __ = _c_int_array(csc.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csc.data)
Guolin Ke's avatar
Guolin Ke committed
1655

1656
        assert csc.shape[0] <= _MAX_INT32
1657
        csc_indices = csc.indices.astype(np.int32, copy=False)
1658

1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
        _safe_call(
            _LIB.LGBM_BoosterPredictForCSC(
                self._handle,
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csc.indptr)),
                ctypes.c_int64(len(csc.data)),
                ctypes.c_int64(csc.shape[0]),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.byref(out_num_preds),
                preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
            )
        )
wxchan's avatar
wxchan committed
1678
        if n_preds != out_num_preds.value:
1679
            raise ValueError("Wrong length for predict results")
wxchan's avatar
wxchan committed
1680
        return preds, nrow
1681

1682
1683
1684
1685
1686
    def __pred_for_pyarrow_table(
        self,
        table: pa_Table,
        start_iteration: int,
        num_iteration: int,
1687
        predict_type: int,
1688
1689
1690
1691
1692
1693
    ) -> Tuple[np.ndarray, int]:
        """Predict for a PyArrow table."""
        if not PYARROW_INSTALLED:
            raise LightGBMError("Cannot predict from Arrow without `pyarrow` installed.")

        # Check that the input is valid: we only handle numbers (for now)
1694
        if not all(arrow_is_integer(t) or arrow_is_floating(t) or arrow_is_boolean(t) for t in table.schema.types):
1695
1696
1697
1698
1699
1700
1701
            raise ValueError("Arrow table may only have integer or floating point datatypes")

        # Prepare prediction output array
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=table.num_rows,
1702
            predict_type=predict_type,
1703
1704
1705
1706
1707
1708
        )
        preds = np.empty(n_preds, dtype=np.float64)
        out_num_preds = ctypes.c_int64(0)

        # Export Arrow table to C and run prediction
        c_array = _export_arrow_to_c(table)
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
        _safe_call(
            _LIB.LGBM_BoosterPredictForArrow(
                self._handle,
                ctypes.c_int64(c_array.n_chunks),
                ctypes.c_void_p(c_array.chunks_ptr),
                ctypes.c_void_p(c_array.schema_ptr),
                ctypes.c_int(predict_type),
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                _c_str(self.pred_parameter),
                ctypes.byref(out_num_preds),
                preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
            )
        )
1723
1724
1725
        if n_preds != out_num_preds.value:
            raise ValueError("Wrong length for predict results")
        return preds, table.num_rows
wxchan's avatar
wxchan committed
1726

1727
    def current_iteration(self) -> int:
1728
1729
1730
1731
1732
1733
1734
1735
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
        out_cur_iter = ctypes.c_int(0)
1736
1737
1738
1739
1740
1741
        _safe_call(
            _LIB.LGBM_BoosterGetCurrentIteration(
                self._handle,
                ctypes.byref(out_cur_iter),
            )
        )
1742
1743
        return out_cur_iter.value

wxchan's avatar
wxchan committed
1744

1745
class Dataset:
wxchan's avatar
wxchan committed
1746
    """Dataset in LightGBM."""
1747

1748
1749
    def __init__(
        self,
1750
        data: _LGBM_TrainDataType,
1751
        label: Optional[_LGBM_LabelType] = None,
1752
        reference: Optional["Dataset"] = None,
1753
1754
1755
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
1756
1757
        feature_name: _LGBM_FeatureNameConfiguration = "auto",
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = "auto",
1758
        params: Optional[Dict[str, Any]] = None,
1759
1760
        free_raw_data: bool = True,
        position: Optional[_LGBM_PositionType] = None,
1761
    ):
1762
        """Initialize Dataset.
1763

wxchan's avatar
wxchan committed
1764
1765
        Parameters
        ----------
1766
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence, list of numpy array or pyarrow Table
wxchan's avatar
wxchan committed
1767
            Data source of Dataset.
1768
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
1769
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1770
1771
1772
            Label of the data.
        reference : Dataset or None, optional (default=None)
            If this is Dataset for validation, training data should be used as reference.
1773
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1774
            Weight for each instance. Weights should be non-negative.
1775
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1776
1777
1778
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
1779
1780
            For example, if you have a 100-document dataset with ``group = [10, 20, 40, 10, 10, 10]``, that means that you have 6 groups,
            where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.
1781
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray, pyarrow Table (for multi-class task) or None, optional (default=None)
1782
            Init score for Dataset.
1783
        feature_name : list of str, or 'auto', optional (default="auto")
1784
            Feature names.
1785
            If 'auto' and data is pandas DataFrame or pyarrow Table, data columns names are used.
1786
        categorical_feature : list of str or int, or 'auto', optional (default="auto")
1787
1788
            Categorical features.
            If list of int, interpreted as indices.
1789
            If list of str, interpreted as feature names (need to specify ``feature_name`` as well).
1790
            If 'auto' and data is pandas DataFrame, pandas unordered categorical columns are used.
1791
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
1792
            Large values could be memory consuming. Consider using consecutive integers starting from zero.
1793
            All negative values in categorical features will be treated as missing values.
1794
            The output cannot be monotonically constrained with respect to a categorical feature.
1795
            Floating point numbers in categorical features will be rounded towards 0.
Nikita Titov's avatar
Nikita Titov committed
1796
        params : dict or None, optional (default=None)
1797
            Other parameters for Dataset.
Nikita Titov's avatar
Nikita Titov committed
1798
        free_raw_data : bool, optional (default=True)
1799
            If True, raw data is freed after constructing inner Dataset.
1800
1801
        position : numpy 1-D array, pandas Series or None, optional (default=None)
            Position of items used in unbiased learning-to-rank task.
wxchan's avatar
wxchan committed
1802
        """
1803
        self._handle: Optional[_DatasetHandle] = None
wxchan's avatar
wxchan committed
1804
1805
1806
1807
1808
        self.data = data
        self.label = label
        self.reference = reference
        self.weight = weight
        self.group = group
1809
        self.position = position
1810
        self.init_score = init_score
1811
1812
        self.feature_name: _LGBM_FeatureNameConfiguration = feature_name
        self.categorical_feature: _LGBM_CategoricalFeatureConfiguration = categorical_feature
1813
        self.params = deepcopy(params)
wxchan's avatar
wxchan committed
1814
        self.free_raw_data = free_raw_data
1815
        self.used_indices: Optional[List[int]] = None
1816
        self._need_slice = True
1817
        self._predictor: Optional[_InnerPredictor] = None
1818
        self.pandas_categorical: Optional[List[List]] = None
1819
        self._params_back_up: Optional[Dict[str, Any]] = None
1820
        self.version = 0
1821
        self._start_row = 0  # Used when pushing rows one by one.
wxchan's avatar
wxchan committed
1822

1823
    def __del__(self) -> None:
1824
1825
1826
1827
        try:
            self._free_handle()
        except AttributeError:
            pass
1828

1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
    def _create_sample_indices(self, total_nrow: int) -> np.ndarray:
        """Get an array of randomly chosen indices from this ``Dataset``.

        Indices are sampled without replacement.

        Parameters
        ----------
        total_nrow : int
            Total number of rows to sample from.
            If this value is greater than the value of parameter ``bin_construct_sample_cnt``, only ``bin_construct_sample_cnt`` indices will be used.
            If Dataset has multiple input data, this should be the sum of rows of every file.

        Returns
        -------
        indices : numpy array
            Indices for sampled data.
        """
1846
        param_str = _param_dict_to_str(self.get_params())
1847
1848
        sample_cnt = _get_sample_count(total_nrow, param_str)
        indices = np.empty(sample_cnt, dtype=np.int32)
1849
        ptr_data, _, _ = _c_int_array(indices)
1850
1851
        actual_sample_cnt = ctypes.c_int32(0)

1852
1853
1854
1855
1856
1857
1858
1859
        _safe_call(
            _LIB.LGBM_SampleIndices(
                ctypes.c_int32(total_nrow),
                _c_str(param_str),
                ptr_data,
                ctypes.byref(actual_sample_cnt),
            )
        )
1860
1861
        assert sample_cnt == actual_sample_cnt.value
        return indices
1862

1863
1864
1865
    def _init_from_ref_dataset(
        self,
        total_nrow: int,
1866
1867
        ref_dataset: _DatasetHandle,
    ) -> "Dataset":
1868
1869
1870
1871
1872
1873
        """Create dataset from a reference dataset.

        Parameters
        ----------
        total_nrow : int
            Number of rows expected to add to dataset.
1874
1875
        ref_dataset : object
            Handle of reference dataset to extract metadata from.
1876
1877
1878
1879
1880
1881

        Returns
        -------
        self : Dataset
            Constructed Dataset object.
        """
1882
        self._handle = ctypes.c_void_p()
1883
1884
1885
1886
1887
1888
1889
        _safe_call(
            _LIB.LGBM_DatasetCreateByReference(
                ref_dataset,
                ctypes.c_int64(total_nrow),
                ctypes.byref(self._handle),
            )
        )
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
        return self

    def _init_from_sample(
        self,
        sample_data: List[np.ndarray],
        sample_indices: List[np.ndarray],
        sample_cnt: int,
        total_nrow: int,
    ) -> "Dataset":
        """Create Dataset from sampled data structures.

        Parameters
        ----------
1903
        sample_data : list of numpy array
1904
            Sample data for each column.
1905
        sample_indices : list of numpy array
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
            Sample data row index for each column.
        sample_cnt : int
            Number of samples.
        total_nrow : int
            Total number of rows for all input files.

        Returns
        -------
        self : Dataset
            Constructed Dataset object.
        """
        ncol = len(sample_indices)
        assert len(sample_data) == ncol, "#sample data column != #column indices"

        for i in range(ncol):
            if sample_data[i].dtype != np.double:
                raise ValueError(f"sample_data[{i}] type {sample_data[i].dtype} is not double")
            if sample_indices[i].dtype != np.int32:
                raise ValueError(f"sample_indices[{i}] type {sample_indices[i].dtype} is not int32")

        # c type: double**
        # each double* element points to start of each column of sample data.
1928
        sample_col_ptr: _ctypes_float_array = (ctypes.POINTER(ctypes.c_double) * ncol)()
1929
1930
        # c type int**
        # each int* points to start of indices for each column
1931
        indices_col_ptr: _ctypes_int_array = (ctypes.POINTER(ctypes.c_int32) * ncol)()
1932
        for i in range(ncol):
1933
1934
            sample_col_ptr[i] = _c_float_array(sample_data[i])[0]
            indices_col_ptr[i] = _c_int_array(sample_indices[i])[0]
1935
1936

        num_per_col = np.array([len(d) for d in sample_indices], dtype=np.int32)
1937
        num_per_col_ptr, _, _ = _c_int_array(num_per_col)
1938

1939
        self._handle = ctypes.c_void_p()
1940
        params_str = _param_dict_to_str(self.get_params())
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
        _safe_call(
            _LIB.LGBM_DatasetCreateFromSampledColumn(
                ctypes.cast(sample_col_ptr, ctypes.POINTER(ctypes.POINTER(ctypes.c_double))),
                ctypes.cast(indices_col_ptr, ctypes.POINTER(ctypes.POINTER(ctypes.c_int32))),
                ctypes.c_int32(ncol),
                num_per_col_ptr,
                ctypes.c_int32(sample_cnt),
                ctypes.c_int32(total_nrow),
                ctypes.c_int64(total_nrow),
                _c_str(params_str),
                ctypes.byref(self._handle),
            )
        )
1954
1955
        return self

1956
    def _push_rows(self, data: np.ndarray) -> "Dataset":
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
        """Add rows to Dataset.

        Parameters
        ----------
        data : numpy 1-D array
            New data to add to the Dataset.

        Returns
        -------
        self : Dataset
            Dataset object.
        """
        nrow, ncol = data.shape
        data = data.reshape(data.size)
1971
        data_ptr, data_type, _ = _c_float_array(data)
1972

1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
        _safe_call(
            _LIB.LGBM_DatasetPushRows(
                self._handle,
                data_ptr,
                data_type,
                ctypes.c_int32(nrow),
                ctypes.c_int32(ncol),
                ctypes.c_int32(self._start_row),
            )
        )
1983
1984
1985
        self._start_row += nrow
        return self

1986
    def get_params(self) -> Dict[str, Any]:
1987
1988
1989
1990
        """Get the used parameters in the Dataset.

        Returns
        -------
1991
        params : dict
1992
1993
1994
1995
            The used parameters in this Dataset object.
        """
        if self.params is not None:
            # no min_data, nthreads and verbose in this function
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
            dataset_params = _ConfigAliases.get(
                "bin_construct_sample_cnt",
                "categorical_feature",
                "data_random_seed",
                "enable_bundle",
                "feature_pre_filter",
                "forcedbins_filename",
                "group_column",
                "header",
                "ignore_column",
                "is_enable_sparse",
                "label_column",
                "linear_tree",
                "max_bin",
                "max_bin_by_feature",
                "min_data_in_bin",
                "pre_partition",
                "precise_float_parser",
                "two_round",
                "use_missing",
                "weight_column",
                "zero_as_missing",
            )
2019
            return {k: v for k, v in self.params.items() if k in dataset_params}
2020
2021
        else:
            return {}
2022

2023
    def _free_handle(self) -> "Dataset":
2024
2025
2026
        if self._handle is not None:
            _safe_call(_LIB.LGBM_DatasetFree(self._handle))
            self._handle = None
2027
        self._need_slice = True
Guolin Ke's avatar
Guolin Ke committed
2028
2029
        if self.used_indices is not None:
            self.data = None
Nikita Titov's avatar
Nikita Titov committed
2030
        return self
wxchan's avatar
wxchan committed
2031

2032
2033
2034
    def _set_init_score_by_predictor(
        self,
        predictor: Optional[_InnerPredictor],
2035
        data: _LGBM_TrainDataType,
2036
        used_indices: Optional[Union[List[int], np.ndarray]],
2037
    ) -> "Dataset":
Guolin Ke's avatar
Guolin Ke committed
2038
        data_has_header = False
2039
        if isinstance(data, (str, Path)) and self.params is not None:
Guolin Ke's avatar
Guolin Ke committed
2040
            # check data has header or not
2041
            data_has_header = any(self.params.get(alias, False) for alias in _ConfigAliases.get("header"))
Guolin Ke's avatar
Guolin Ke committed
2042
        num_data = self.num_data()
2043
        if predictor is not None:
2044
2045
2046
            init_score: Union[np.ndarray, scipy.sparse.spmatrix] = predictor.predict(
                data=data,
                raw_score=True,
2047
                data_has_header=data_has_header,
2048
            )
2049
            init_score = init_score.ravel()
2050
            if used_indices is not None:
2051
                assert not self._need_slice
2052
                if isinstance(data, (str, Path)):
2053
                    sub_init_score = np.empty(num_data * predictor.num_class, dtype=np.float64)
2054
                    assert num_data == len(used_indices)
2055
2056
                    for i in range(len(used_indices)):
                        for j in range(predictor.num_class):
2057
2058
2059
                            sub_init_score[i * predictor.num_class + j] = init_score[
                                used_indices[i] * predictor.num_class + j
                            ]
2060
2061
2062
                    init_score = sub_init_score
            if predictor.num_class > 1:
                # need to regroup init_score
2063
                new_init_score = np.empty(init_score.size, dtype=np.float64)
2064
2065
                for i in range(num_data):
                    for j in range(predictor.num_class):
2066
2067
2068
                        new_init_score[j * num_data + i] = init_score[i * predictor.num_class + j]
                init_score = new_init_score
        elif self.init_score is not None:
2069
            init_score = np.full_like(self.init_score, fill_value=0.0, dtype=np.float64)
2070
2071
        else:
            return self
Guolin Ke's avatar
Guolin Ke committed
2072
        self.set_init_score(init_score)
2073
        return self
Guolin Ke's avatar
Guolin Ke committed
2074

2075
2076
    def _lazy_init(
        self,
2077
        data: Optional[_LGBM_TrainDataType],
2078
2079
2080
2081
2082
2083
2084
2085
        label: Optional[_LGBM_LabelType],
        reference: Optional["Dataset"],
        weight: Optional[_LGBM_WeightType],
        group: Optional[_LGBM_GroupType],
        init_score: Optional[_LGBM_InitScoreType],
        predictor: Optional[_InnerPredictor],
        feature_name: _LGBM_FeatureNameConfiguration,
        categorical_feature: _LGBM_CategoricalFeatureConfiguration,
2086
        params: Optional[Dict[str, Any]],
2087
        position: Optional[_LGBM_PositionType],
2088
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2089
        if data is None:
2090
            self._handle = None
Nikita Titov's avatar
Nikita Titov committed
2091
            return self
Guolin Ke's avatar
Guolin Ke committed
2092
2093
2094
        if reference is not None:
            self.pandas_categorical = reference.pandas_categorical
            categorical_feature = reference.categorical_feature
2095
2096
2097
2098
2099
        if isinstance(data, pd_DataFrame):
            data, feature_name, categorical_feature, self.pandas_categorical = _data_from_pandas(
                data=data,
                feature_name=feature_name,
                categorical_feature=categorical_feature,
2100
                pandas_categorical=self.pandas_categorical,
2101
            )
Guolin Ke's avatar
Guolin Ke committed
2102

2103
        # process for args
wxchan's avatar
wxchan committed
2104
        params = {} if params is None else params
2105
        args_names = inspect.signature(self.__class__._lazy_init).parameters.keys()
2106
        for key in params.keys():
2107
            if key in args_names:
2108
2109
2110
2111
                _log_warning(
                    f"{key} keyword has been found in `params` and will be ignored.\n"
                    f"Please use {key} argument of the Dataset constructor to pass this parameter."
                )
2112
        # get categorical features
2113
        if isinstance(categorical_feature, list):
2114
2115
            categorical_indices = set()
            feature_dict = {}
2116
            if isinstance(feature_name, list):
2117
2118
                feature_dict = {name: i for i, name in enumerate(feature_name)}
            for name in categorical_feature:
2119
                if isinstance(name, str) and name in feature_dict:
2120
                    categorical_indices.add(feature_dict[name])
2121
                elif isinstance(name, int):
2122
2123
                    categorical_indices.add(name)
                else:
2124
                    raise TypeError(f"Wrong type({type(name).__name__}) or unknown name({name}) in categorical_feature")
2125
            if categorical_indices:
2126
2127
                for cat_alias in _ConfigAliases.get("categorical_feature"):
                    if cat_alias in params:
2128
                        # If the params[cat_alias] is equal to categorical_indices, do not report the warning.
2129
                        if not (isinstance(params[cat_alias], list) and set(params[cat_alias]) == categorical_indices):
2130
                            _log_warning(f"{cat_alias} in param dict is overridden.")
2131
                        params.pop(cat_alias, None)
2132
                params["categorical_column"] = sorted(categorical_indices)
2133

2134
        params_str = _param_dict_to_str(params)
2135
        self.params = params
2136
        # process for reference dataset
wxchan's avatar
wxchan committed
2137
        ref_dataset = None
wxchan's avatar
wxchan committed
2138
        if isinstance(reference, Dataset):
2139
            ref_dataset = reference.construct()._handle
wxchan's avatar
wxchan committed
2140
        elif reference is not None:
2141
            raise TypeError("Reference dataset should be None or dataset instance")
2142
        # start construct data
2143
        if isinstance(data, (str, Path)):
2144
            self._handle = ctypes.c_void_p()
2145
2146
2147
2148
2149
2150
2151
2152
            _safe_call(
                _LIB.LGBM_DatasetCreateFromFile(
                    _c_str(str(data)),
                    _c_str(params_str),
                    ref_dataset,
                    ctypes.byref(self._handle),
                )
            )
wxchan's avatar
wxchan committed
2153
2154
        elif isinstance(data, scipy.sparse.csr_matrix):
            self.__init_from_csr(data, params_str, ref_dataset)
Guolin Ke's avatar
Guolin Ke committed
2155
2156
        elif isinstance(data, scipy.sparse.csc_matrix):
            self.__init_from_csc(data, params_str, ref_dataset)
wxchan's avatar
wxchan committed
2157
2158
        elif isinstance(data, np.ndarray):
            self.__init_from_np2d(data, params_str, ref_dataset)
2159
2160
2161
        elif _is_pyarrow_table(data):
            self.__init_from_pyarrow_table(data, params_str, ref_dataset)
            feature_name = data.column_names
2162
        elif isinstance(data, list) and len(data) > 0:
2163
            if _is_list_of_numpy_arrays(data):
2164
                self.__init_from_list_np2d(data, params_str, ref_dataset)
2165
            elif _is_list_of_sequences(data):
2166
2167
                self.__init_from_seqs(data, ref_dataset)
            else:
2168
                raise TypeError("Data list can only be of ndarray or Sequence")
2169
2170
        elif isinstance(data, Sequence):
            self.__init_from_seqs([data], ref_dataset)
2171
        elif isinstance(data, dt_DataTable):
2172
            self.__init_from_np2d(data.to_numpy(), params_str, ref_dataset)
wxchan's avatar
wxchan committed
2173
2174
2175
2176
        else:
            try:
                csr = scipy.sparse.csr_matrix(data)
                self.__init_from_csr(csr, params_str, ref_dataset)
2177
            except BaseException as err:
2178
                raise TypeError(f"Cannot initialize Dataset from {type(data).__name__}") from err
wxchan's avatar
wxchan committed
2179
2180
2181
        if label is not None:
            self.set_label(label)
        if self.get_label() is None:
2182
            raise ValueError("Label should not be None")
wxchan's avatar
wxchan committed
2183
2184
2185
2186
        if weight is not None:
            self.set_weight(weight)
        if group is not None:
            self.set_group(group)
2187
2188
        if position is not None:
            self.set_position(position)
2189
2190
        if isinstance(predictor, _InnerPredictor):
            if self._predictor is None and init_score is not None:
2191
                _log_warning("The init_score will be overridden by the prediction of init_model.")
2192
            self._set_init_score_by_predictor(predictor=predictor, data=data, used_indices=None)
2193
2194
        elif init_score is not None:
            self.set_init_score(init_score)
Guolin Ke's avatar
Guolin Ke committed
2195
        elif predictor is not None:
2196
            raise TypeError(f"Wrong predictor type {type(predictor).__name__}")
Guolin Ke's avatar
Guolin Ke committed
2197
        # set feature names
Nikita Titov's avatar
Nikita Titov committed
2198
        return self.set_feature_name(feature_name)
wxchan's avatar
wxchan committed
2199

2200
    @staticmethod
2201
    def _yield_row_from_seqlist(seqs: List[Sequence], indices: Iterable[int]) -> Iterator[np.ndarray]:
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
        offset = 0
        seq_id = 0
        seq = seqs[seq_id]
        for row_id in indices:
            assert row_id >= offset, "sample indices are expected to be monotonic"
            while row_id >= offset + len(seq):
                offset += len(seq)
                seq_id += 1
                seq = seqs[seq_id]
            id_in_seq = row_id - offset
            row = seq[id_in_seq]
2213
            yield row if row.flags["OWNDATA"] else row.copy()
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226

    def __sample(self, seqs: List[Sequence], total_nrow: int) -> Tuple[List[np.ndarray], List[np.ndarray]]:
        """Sample data from seqs.

        Mimics behavior in c_api.cpp:LGBM_DatasetCreateFromMats()

        Returns
        -------
            sampled_rows, sampled_row_indices
        """
        indices = self._create_sample_indices(total_nrow)

        # Select sampled rows, transpose to column order.
2227
        sampled = np.array(list(self._yield_row_from_seqlist(seqs, indices)))
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
        sampled = sampled.T

        filtered = []
        filtered_idx = []
        sampled_row_range = np.arange(len(indices), dtype=np.int32)
        for col in sampled:
            col_predicate = (np.abs(col) > ZERO_THRESHOLD) | np.isnan(col)
            filtered_col = col[col_predicate]
            filtered_row_idx = sampled_row_range[col_predicate]

            filtered.append(filtered_col)
            filtered_idx.append(filtered_row_idx)

        return filtered, filtered_idx

2243
2244
2245
    def __init_from_seqs(
        self,
        seqs: List[Sequence],
2246
        ref_dataset: Optional[_DatasetHandle],
2247
    ) -> "Dataset":
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
        """
        Initialize data from list of Sequence objects.

        Sequence: Generic Data Access Object
            Supports random access and access by batch if properly defined by user

        Data scheme uniformity are trusted, not checked
        """
        total_nrow = sum(len(seq) for seq in seqs)

        # create validation dataset from ref_dataset
        if ref_dataset is not None:
            self._init_from_ref_dataset(total_nrow, ref_dataset)
        else:
2262
            param_str = _param_dict_to_str(self.get_params())
2263
2264
2265
2266
2267
2268
2269
            sample_cnt = _get_sample_count(total_nrow, param_str)

            sample_data, col_indices = self.__sample(seqs, total_nrow)
            self._init_from_sample(sample_data, col_indices, sample_cnt, total_nrow)

        for seq in seqs:
            nrow = len(seq)
2270
            batch_size = getattr(seq, "batch_size", None) or Sequence.batch_size
2271
2272
2273
2274
2275
            for start in range(0, nrow, batch_size):
                end = min(start + batch_size, nrow)
                self._push_rows(seq[start:end])
        return self

2276
2277
2278
2279
    def __init_from_np2d(
        self,
        mat: np.ndarray,
        params_str: str,
2280
        ref_dataset: Optional[_DatasetHandle],
2281
    ) -> "Dataset":
2282
        """Initialize data from a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
2283
        if len(mat.shape) != 2:
2284
            raise ValueError("Input numpy.ndarray must be 2 dimensional")
wxchan's avatar
wxchan committed
2285

2286
        self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2287
2288
        if mat.dtype == np.float32 or mat.dtype == np.float64:
            data = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
2289
        else:  # change non-float data to float data, need to copy
wxchan's avatar
wxchan committed
2290
2291
            data = np.array(mat.reshape(mat.size), dtype=np.float32)

2292
        ptr_data, type_ptr_data, _ = _c_float_array(data)
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
        _safe_call(
            _LIB.LGBM_DatasetCreateFromMat(
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int32(mat.shape[0]),
                ctypes.c_int32(mat.shape[1]),
                ctypes.c_int(_C_API_IS_ROW_MAJOR),
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2305
        return self
wxchan's avatar
wxchan committed
2306

2307
2308
2309
2310
    def __init_from_list_np2d(
        self,
        mats: List[np.ndarray],
        params_str: str,
2311
        ref_dataset: Optional[_DatasetHandle],
2312
    ) -> "Dataset":
2313
        """Initialize data from a list of 2-D numpy matrices."""
2314
        ncol = mats[0].shape[1]
2315
        nrow = np.empty((len(mats),), np.int32)
2316
        ptr_data: _ctypes_float_array
2317
2318
2319
2320
2321
2322
        if mats[0].dtype == np.float64:
            ptr_data = (ctypes.POINTER(ctypes.c_double) * len(mats))()
        else:
            ptr_data = (ctypes.POINTER(ctypes.c_float) * len(mats))()

        holders = []
2323
        type_ptr_data = -1
2324
2325
2326

        for i, mat in enumerate(mats):
            if len(mat.shape) != 2:
2327
                raise ValueError("Input numpy.ndarray must be 2 dimensional")
2328
2329

            if mat.shape[1] != ncol:
2330
                raise ValueError("Input arrays must have same number of columns")
2331
2332
2333
2334
2335

            nrow[i] = mat.shape[0]

            if mat.dtype == np.float32 or mat.dtype == np.float64:
                mats[i] = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
2336
            else:  # change non-float data to float data, need to copy
2337
2338
                mats[i] = np.array(mat.reshape(mat.size), dtype=np.float32)

2339
            chunk_ptr_data, chunk_type_ptr_data, holder = _c_float_array(mats[i])
2340
            if type_ptr_data != -1 and chunk_type_ptr_data != type_ptr_data:
2341
                raise ValueError("Input chunks must have same type")
2342
2343
2344
2345
            ptr_data[i] = chunk_ptr_data
            type_ptr_data = chunk_type_ptr_data
            holders.append(holder)

2346
        self._handle = ctypes.c_void_p()
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
        _safe_call(
            _LIB.LGBM_DatasetCreateFromMats(
                ctypes.c_int32(len(mats)),
                ctypes.cast(ptr_data, ctypes.POINTER(ctypes.POINTER(ctypes.c_double))),
                ctypes.c_int(type_ptr_data),
                nrow.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ctypes.c_int32(ncol),
                ctypes.c_int(_C_API_IS_ROW_MAJOR),
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2360
        return self
2361

2362
2363
2364
2365
    def __init_from_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        params_str: str,
2366
        ref_dataset: Optional[_DatasetHandle],
2367
    ) -> "Dataset":
2368
        """Initialize data from a CSR matrix."""
wxchan's avatar
wxchan committed
2369
        if len(csr.indices) != len(csr.data):
2370
            raise ValueError(f"Length mismatch: {len(csr.indices)} vs {len(csr.data)}")
2371
        self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2372

2373
2374
        ptr_indptr, type_ptr_indptr, __ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
wxchan's avatar
wxchan committed
2375

2376
        assert csr.shape[1] <= _MAX_INT32
2377
        csr_indices = csr.indices.astype(np.int32, copy=False)
2378

2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
        _safe_call(
            _LIB.LGBM_DatasetCreateFromCSR(
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csr.indptr)),
                ctypes.c_int64(len(csr.data)),
                ctypes.c_int64(csr.shape[1]),
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2394
        return self
wxchan's avatar
wxchan committed
2395

2396
2397
2398
2399
    def __init_from_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        params_str: str,
2400
        ref_dataset: Optional[_DatasetHandle],
2401
    ) -> "Dataset":
2402
        """Initialize data from a CSC matrix."""
Guolin Ke's avatar
Guolin Ke committed
2403
        if len(csc.indices) != len(csc.data):
2404
            raise ValueError(f"Length mismatch: {len(csc.indices)} vs {len(csc.data)}")
2405
        self._handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
2406

2407
2408
        ptr_indptr, type_ptr_indptr, __ = _c_int_array(csc.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csc.data)
Guolin Ke's avatar
Guolin Ke committed
2409

2410
        assert csc.shape[0] <= _MAX_INT32
2411
        csc_indices = csc.indices.astype(np.int32, copy=False)
2412

2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
        _safe_call(
            _LIB.LGBM_DatasetCreateFromCSC(
                ptr_indptr,
                ctypes.c_int(type_ptr_indptr),
                csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                ptr_data,
                ctypes.c_int(type_ptr_data),
                ctypes.c_int64(len(csc.indptr)),
                ctypes.c_int64(len(csc.data)),
                ctypes.c_int64(csc.shape[0]),
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2428
        return self
Guolin Ke's avatar
Guolin Ke committed
2429

2430
2431
2432
2433
    def __init_from_pyarrow_table(
        self,
        table: pa_Table,
        params_str: str,
2434
        ref_dataset: Optional[_DatasetHandle],
2435
2436
2437
2438
2439
2440
    ) -> "Dataset":
        """Initialize data from a PyArrow table."""
        if not PYARROW_INSTALLED:
            raise LightGBMError("Cannot init dataframe from Arrow without `pyarrow` installed.")

        # Check that the input is valid: we only handle numbers (for now)
2441
        if not all(arrow_is_integer(t) or arrow_is_floating(t) or arrow_is_boolean(t) for t in table.schema.types):
2442
2443
2444
2445
2446
            raise ValueError("Arrow table may only have integer or floating point datatypes")

        # Export Arrow table to C
        c_array = _export_arrow_to_c(table)
        self._handle = ctypes.c_void_p()
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
        _safe_call(
            _LIB.LGBM_DatasetCreateFromArrow(
                ctypes.c_int64(c_array.n_chunks),
                ctypes.c_void_p(c_array.chunks_ptr),
                ctypes.c_void_p(c_array.schema_ptr),
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
2457
2458
        return self

2459
    @staticmethod
2460
    def _compare_params_for_warning(
2461
2462
        params: Dict[str, Any],
        other_params: Dict[str, Any],
2463
        ignore_keys: Set[str],
2464
2465
    ) -> bool:
        """Compare two dictionaries with params ignoring some keys.
2466

2467
2468
2469
2470
        It is only for the warning purpose.

        Parameters
        ----------
2471
        params : dict
2472
            One dictionary with parameters to compare.
2473
        other_params : dict
2474
2475
2476
            Another dictionary with parameters to compare.
        ignore_keys : set
            Keys that should be ignored during comparing two dictionaries.
2477
2478
2479

        Returns
        -------
2480
2481
        compare_result : bool
          Returns whether two dictionaries with params are equal.
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
        """
        for k in other_params:
            if k not in ignore_keys:
                if k not in params or params[k] != other_params[k]:
                    return False
        for k in params:
            if k not in ignore_keys:
                if k not in other_params or params[k] != other_params[k]:
                    return False
        return True

2493
    def construct(self) -> "Dataset":
2494
2495
2496
2497
2498
        """Lazy init.

        Returns
        -------
        self : Dataset
Nikita Titov's avatar
Nikita Titov committed
2499
            Constructed Dataset object.
2500
        """
2501
        if self._handle is None:
wxchan's avatar
wxchan committed
2502
            if self.reference is not None:
2503
                reference_params = self.reference.get_params()
2504
2505
                params = self.get_params()
                if params != reference_params:
2506
2507
2508
                    if not self._compare_params_for_warning(
                        params=params,
                        other_params=reference_params,
2509
                        ignore_keys=_ConfigAliases.get("categorical_feature"),
2510
                    ):
2511
                        _log_warning("Overriding the parameters from Reference Dataset.")
2512
                    self._update_params(reference_params)
wxchan's avatar
wxchan committed
2513
                if self.used_indices is None:
2514
                    # create valid
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
                    self._lazy_init(
                        data=self.data,
                        label=self.label,
                        reference=self.reference,
                        weight=self.weight,
                        group=self.group,
                        position=self.position,
                        init_score=self.init_score,
                        predictor=self._predictor,
                        feature_name=self.feature_name,
                        categorical_feature="auto",
                        params=self.params,
                    )
wxchan's avatar
wxchan committed
2528
                else:
2529
                    # construct subset
2530
                    used_indices = _list_to_1d_numpy(self.used_indices, dtype=np.int32, name="used_indices")
2531
                    assert used_indices.flags.c_contiguous
Guolin Ke's avatar
Guolin Ke committed
2532
                    if self.reference.group is not None:
2533
                        group_info = np.array(self.reference.group).astype(np.int32, copy=False)
2534
2535
2536
                        _, self.group = np.unique(
                            np.repeat(range(len(group_info)), repeats=group_info)[self.used_indices], return_counts=True
                        )
2537
                    self._handle = ctypes.c_void_p()
2538
                    params_str = _param_dict_to_str(self.params)
2539
2540
2541
2542
2543
2544
2545
2546
2547
                    _safe_call(
                        _LIB.LGBM_DatasetGetSubset(
                            self.reference.construct()._handle,
                            used_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
                            ctypes.c_int32(used_indices.shape[0]),
                            _c_str(params_str),
                            ctypes.byref(self._handle),
                        )
                    )
Guolin Ke's avatar
Guolin Ke committed
2548
2549
                    if not self.free_raw_data:
                        self.get_data()
Guolin Ke's avatar
Guolin Ke committed
2550
2551
                    if self.group is not None:
                        self.set_group(self.group)
2552
2553
                    if self.position is not None:
                        self.set_position(self.position)
wxchan's avatar
wxchan committed
2554
2555
                    if self.get_label() is None:
                        raise ValueError("Label should not be None.")
2556
2557
2558
2559
                    if (
                        isinstance(self._predictor, _InnerPredictor)
                        and self._predictor is not self.reference._predictor
                    ):
Guolin Ke's avatar
Guolin Ke committed
2560
                        self.get_data()
2561
                        self._set_init_score_by_predictor(
2562
                            predictor=self._predictor, data=self.data, used_indices=used_indices
2563
                        )
wxchan's avatar
wxchan committed
2564
            else:
2565
                # create train
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
                self._lazy_init(
                    data=self.data,
                    label=self.label,
                    reference=None,
                    weight=self.weight,
                    group=self.group,
                    init_score=self.init_score,
                    predictor=self._predictor,
                    feature_name=self.feature_name,
                    categorical_feature=self.categorical_feature,
                    params=self.params,
                    position=self.position,
                )
wxchan's avatar
wxchan committed
2579
2580
            if self.free_raw_data:
                self.data = None
2581
            self.feature_name = self.get_feature_name()
wxchan's avatar
wxchan committed
2582
        return self
wxchan's avatar
wxchan committed
2583

2584
2585
    def create_valid(
        self,
2586
        data: _LGBM_TrainDataType,
2587
        label: Optional[_LGBM_LabelType] = None,
2588
2589
2590
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
2591
        params: Optional[Dict[str, Any]] = None,
2592
        position: Optional[_LGBM_PositionType] = None,
2593
    ) -> "Dataset":
2594
        """Create validation data align with current Dataset.
wxchan's avatar
wxchan committed
2595
2596
2597

        Parameters
        ----------
2598
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array
wxchan's avatar
wxchan committed
2599
            Data source of Dataset.
2600
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
2601
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2602
            Label of the data.
2603
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2604
            Weight for each instance. Weights should be non-negative.
2605
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2606
2607
2608
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2609
2610
            For example, if you have a 100-document dataset with ``group = [10, 20, 40, 10, 10, 10]``, that means that you have 6 groups,
            where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.
2611
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray, pyarrow Table (for multi-class task) or None, optional (default=None)
2612
            Init score for Dataset.
Nikita Titov's avatar
Nikita Titov committed
2613
        params : dict or None, optional (default=None)
2614
            Other parameters for validation Dataset.
2615
2616
        position : numpy 1-D array, pandas Series or None, optional (default=None)
            Position of items used in unbiased learning-to-rank task.
2617
2618
2619

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2620
2621
        valid : Dataset
            Validation Dataset with reference to self.
wxchan's avatar
wxchan committed
2622
        """
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
        ret = Dataset(
            data,
            label=label,
            reference=self,
            weight=weight,
            group=group,
            position=position,
            init_score=init_score,
            params=params,
            free_raw_data=self.free_raw_data,
        )
wxchan's avatar
wxchan committed
2634
        ret._predictor = self._predictor
2635
        ret.pandas_categorical = self.pandas_categorical
wxchan's avatar
wxchan committed
2636
        return ret
wxchan's avatar
wxchan committed
2637

2638
2639
2640
    def subset(
        self,
        used_indices: List[int],
2641
        params: Optional[Dict[str, Any]] = None,
2642
    ) -> "Dataset":
2643
        """Get subset of current Dataset.
wxchan's avatar
wxchan committed
2644
2645
2646
2647

        Parameters
        ----------
        used_indices : list of int
2648
            Indices used to create the subset.
Nikita Titov's avatar
Nikita Titov committed
2649
        params : dict or None, optional (default=None)
2650
            These parameters will be passed to Dataset constructor.
2651
2652
2653
2654
2655

        Returns
        -------
        subset : Dataset
            Subset of the current Dataset.
wxchan's avatar
wxchan committed
2656
        """
wxchan's avatar
wxchan committed
2657
2658
        if params is None:
            params = self.params
2659
2660
2661
2662
2663
2664
2665
2666
        ret = Dataset(
            None,
            reference=self,
            feature_name=self.feature_name,
            categorical_feature=self.categorical_feature,
            params=params,
            free_raw_data=self.free_raw_data,
        )
wxchan's avatar
wxchan committed
2667
        ret._predictor = self._predictor
2668
        ret.pandas_categorical = self.pandas_categorical
2669
        ret.used_indices = sorted(used_indices)
wxchan's avatar
wxchan committed
2670
2671
        return ret

2672
    def save_binary(self, filename: Union[str, Path]) -> "Dataset":
2673
        """Save Dataset to a binary file.
wxchan's avatar
wxchan committed
2674

2675
2676
2677
2678
2679
        .. note::

            Please note that `init_score` is not saved in binary file.
            If you need it, please set it again after loading Dataset.

wxchan's avatar
wxchan committed
2680
2681
        Parameters
        ----------
2682
        filename : str or pathlib.Path
wxchan's avatar
wxchan committed
2683
            Name of the output file.
Nikita Titov's avatar
Nikita Titov committed
2684
2685
2686
2687
2688

        Returns
        -------
        self : Dataset
            Returns self.
wxchan's avatar
wxchan committed
2689
        """
2690
2691
2692
2693
2694
2695
        _safe_call(
            _LIB.LGBM_DatasetSaveBinary(
                self.construct()._handle,
                _c_str(str(filename)),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2696
        return self
wxchan's avatar
wxchan committed
2697

2698
    def _update_params(self, params: Optional[Dict[str, Any]]) -> "Dataset":
2699
2700
        if not params:
            return self
2701
        params = deepcopy(params)
2702

2703
        def update() -> None:
2704
2705
2706
            if not self.params:
                self.params = params
            else:
2707
                self._params_back_up = deepcopy(self.params)
2708
2709
                self.params.update(params)

2710
        if self._handle is None:
2711
2712
2713
            update()
        elif params is not None:
            ret = _LIB.LGBM_DatasetUpdateParamChecking(
2714
                _c_str(_param_dict_to_str(self.params)),
2715
2716
                _c_str(_param_dict_to_str(params)),
            )
2717
2718
2719
2720
2721
2722
            if ret != 0:
                # could be updated if data is not freed
                if self.data is not None:
                    update()
                    self._free_handle()
                else:
2723
                    raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
Nikita Titov's avatar
Nikita Titov committed
2724
        return self
wxchan's avatar
wxchan committed
2725

2726
    def _reverse_update_params(self) -> "Dataset":
2727
        if self._handle is None:
2728
2729
            self.params = deepcopy(self._params_back_up)
            self._params_back_up = None
Nikita Titov's avatar
Nikita Titov committed
2730
        return self
2731

2732
2733
2734
    def set_field(
        self,
        field_name: str,
2735
        data: Optional[_LGBM_SetFieldType],
2736
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2737
        """Set property into the Dataset.
wxchan's avatar
wxchan committed
2738
2739
2740

        Parameters
        ----------
2741
        field_name : str
2742
            The field name of the information.
2743
        data : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray or None
2744
            The data to be set.
Nikita Titov's avatar
Nikita Titov committed
2745
2746
2747
2748
2749

        Returns
        -------
        self : Dataset
            Dataset with set property.
wxchan's avatar
wxchan committed
2750
        """
2751
        if self._handle is None:
2752
            raise Exception(f"Cannot set {field_name} before construct dataset")
wxchan's avatar
wxchan committed
2753
        if data is None:
2754
            # set to None
2755
2756
2757
2758
2759
2760
2761
2762
2763
            _safe_call(
                _LIB.LGBM_DatasetSetField(
                    self._handle,
                    _c_str(field_name),
                    None,
                    ctypes.c_int(0),
                    ctypes.c_int(_FIELD_TYPE_MAPPER[field_name]),
                )
            )
Nikita Titov's avatar
Nikita Titov committed
2764
            return self
2765
2766

        # If the data is a arrow data, we can just pass it to C
2767
2768
2769
2770
2771
2772
        if _is_pyarrow_array(data) or _is_pyarrow_table(data):
            # If a table is being passed, we concatenate the columns. This is only valid for
            # 'init_score'.
            if _is_pyarrow_table(data):
                if field_name != "init_score":
                    raise ValueError(f"pyarrow tables are not supported for field '{field_name}'")
2773
2774
2775
2776
2777
2778
2779
                data = pa_chunked_array(
                    [
                        chunk
                        for array in data.columns  # type: ignore
                        for chunk in array.chunks
                    ]
                )
2780

2781
            c_array = _export_arrow_to_c(data)
2782
2783
2784
2785
2786
2787
2788
2789
2790
            _safe_call(
                _LIB.LGBM_DatasetSetFieldFromArrow(
                    self._handle,
                    _c_str(field_name),
                    ctypes.c_int64(c_array.n_chunks),
                    ctypes.c_void_p(c_array.chunks_ptr),
                    ctypes.c_void_p(c_array.schema_ptr),
                )
            )
2791
2792
2793
            self.version += 1
            return self

2794
        dtype: "np.typing.DTypeLike"
2795
        if field_name == "init_score":
Guolin Ke's avatar
Guolin Ke committed
2796
            dtype = np.float64
2797
            if _is_1d_collection(data):
2798
                data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2799
            elif _is_2d_collection(data):
2800
                data = _data_to_2d_numpy(data, dtype=dtype, name=field_name)
2801
                data = data.ravel(order="F")
2802
2803
            else:
                raise TypeError(
2804
2805
                    "init_score must be list, numpy 1-D array or pandas Series.\n"
                    "In multiclass classification init_score can also be a list of lists, numpy 2-D array or pandas DataFrame."
2806
2807
                )
        else:
2808
            if field_name in {"group", "position"}:
2809
2810
2811
                dtype = np.int32
            else:
                dtype = np.float32
2812
            data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2813

2814
        ptr_data: Union[_ctypes_float_ptr, _ctypes_int_ptr]
2815
        if data.dtype == np.float32 or data.dtype == np.float64:
2816
            ptr_data, type_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
2817
        elif data.dtype == np.int32:
2818
            ptr_data, type_data, _ = _c_int_array(data)
wxchan's avatar
wxchan committed
2819
        else:
2820
            raise TypeError(f"Expected np.float32/64 or np.int32, met type({data.dtype})")
2821
        if type_data != _FIELD_TYPE_MAPPER[field_name]:
2822
            raise TypeError("Input type error for set_field")
2823
2824
2825
2826
2827
2828
2829
2830
2831
        _safe_call(
            _LIB.LGBM_DatasetSetField(
                self._handle,
                _c_str(field_name),
                ptr_data,
                ctypes.c_int(len(data)),
                ctypes.c_int(type_data),
            )
        )
2832
        self.version += 1
Nikita Titov's avatar
Nikita Titov committed
2833
        return self
wxchan's avatar
wxchan committed
2834

2835
    def get_field(self, field_name: str) -> Optional[np.ndarray]:
wxchan's avatar
wxchan committed
2836
        """Get property from the Dataset.
wxchan's avatar
wxchan committed
2837

2838
2839
2840
2841
2842
2843
        Can only be run on a constructed Dataset.

        Unlike ``get_group()``, ``get_init_score()``, ``get_label()``, ``get_position()``, and ``get_weight()``,
        this method ignores any raw data passed into ``lgb.Dataset()`` on the Python side, and will only read
        data from the constructed C++ ``Dataset`` object.

wxchan's avatar
wxchan committed
2844
2845
        Parameters
        ----------
2846
        field_name : str
2847
            The field name of the information.
wxchan's avatar
wxchan committed
2848
2849
2850

        Returns
        -------
2851
        info : numpy array or None
2852
            A numpy array with information from the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2853
        """
2854
        if self._handle is None:
2855
            raise Exception(f"Cannot get {field_name} before construct Dataset")
2856
2857
        tmp_out_len = ctypes.c_int(0)
        out_type = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2858
        ret = ctypes.POINTER(ctypes.c_void_p)()
2859
2860
2861
2862
2863
2864
2865
2866
2867
        _safe_call(
            _LIB.LGBM_DatasetGetField(
                self._handle,
                _c_str(field_name),
                ctypes.byref(tmp_out_len),
                ctypes.byref(ret),
                ctypes.byref(out_type),
            )
        )
2868
        if out_type.value != _FIELD_TYPE_MAPPER[field_name]:
wxchan's avatar
wxchan committed
2869
2870
2871
            raise TypeError("Return type error for get_field")
        if tmp_out_len.value == 0:
            return None
2872
        if out_type.value == _C_API_DTYPE_INT32:
2873
2874
            arr = _cint32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_int32)),
2875
                length=tmp_out_len.value,
2876
            )
2877
        elif out_type.value == _C_API_DTYPE_FLOAT32:
2878
2879
            arr = _cfloat32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_float)),
2880
                length=tmp_out_len.value,
2881
            )
2882
        elif out_type.value == _C_API_DTYPE_FLOAT64:
2883
2884
            arr = _cfloat64_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_double)),
2885
                length=tmp_out_len.value,
2886
            )
2887
        else:
wxchan's avatar
wxchan committed
2888
            raise TypeError("Unknown type")
2889
        if field_name == "init_score":
2890
2891
2892
            num_data = self.num_data()
            num_classes = arr.size // num_data
            if num_classes > 1:
2893
                arr = arr.reshape((num_data, num_classes), order="F")
2894
        return arr
Guolin Ke's avatar
Guolin Ke committed
2895

2896
2897
    def set_categorical_feature(
        self,
2898
        categorical_feature: _LGBM_CategoricalFeatureConfiguration,
2899
    ) -> "Dataset":
2900
        """Set categorical features.
2901
2902
2903

        Parameters
        ----------
2904
        categorical_feature : list of str or int, or 'auto'
2905
            Names or indices of categorical features.
Nikita Titov's avatar
Nikita Titov committed
2906
2907
2908
2909
2910

        Returns
        -------
        self : Dataset
            Dataset with set categorical features.
2911
2912
        """
        if self.categorical_feature == categorical_feature:
Nikita Titov's avatar
Nikita Titov committed
2913
            return self
2914
        if self.data is not None:
2915
2916
            if self.categorical_feature is None:
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2917
                return self._free_handle()
2918
            elif categorical_feature == "auto":
Nikita Titov's avatar
Nikita Titov committed
2919
                return self
2920
            else:
2921
2922
2923
2924
2925
                if self.categorical_feature != "auto":
                    _log_warning(
                        "categorical_feature in Dataset is overridden.\n"
                        f"New categorical_feature is {list(categorical_feature)}"
                    )
2926
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2927
                return self._free_handle()
2928
        else:
2929
2930
2931
2932
            raise LightGBMError(
                "Cannot set categorical feature after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
2933

2934
2935
    def _set_predictor(
        self,
2936
        predictor: Optional[_InnerPredictor],
2937
    ) -> "Dataset":
2938
2939
2940
2941
        """Set predictor for continued training.

        It is not recommended for user to call this function.
        Please use init_model argument in engine.train() or engine.cv() instead.
Guolin Ke's avatar
Guolin Ke committed
2942
        """
2943
        if predictor is None and self._predictor is None:
Nikita Titov's avatar
Nikita Titov committed
2944
            return self
2945
        elif isinstance(predictor, _InnerPredictor) and isinstance(self._predictor, _InnerPredictor):
2946
2947
2948
            if (predictor == self._predictor) and (
                predictor.current_iteration() == self._predictor.current_iteration()
            ):
2949
                return self
2950
        if self._handle is None:
Guolin Ke's avatar
Guolin Ke committed
2951
            self._predictor = predictor
2952
2953
        elif self.data is not None:
            self._predictor = predictor
2954
2955
2956
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.data,
2957
                used_indices=None,
2958
            )
2959
2960
        elif self.used_indices is not None and self.reference is not None and self.reference.data is not None:
            self._predictor = predictor
2961
2962
2963
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.reference.data,
2964
                used_indices=self.used_indices,
2965
            )
Guolin Ke's avatar
Guolin Ke committed
2966
        else:
2967
2968
2969
2970
            raise LightGBMError(
                "Cannot set predictor after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
2971
        return self
Guolin Ke's avatar
Guolin Ke committed
2972

2973
    def set_reference(self, reference: "Dataset") -> "Dataset":
2974
        """Set reference Dataset.
Guolin Ke's avatar
Guolin Ke committed
2975
2976
2977
2978

        Parameters
        ----------
        reference : Dataset
2979
            Reference that is used as a template to construct the current Dataset.
Nikita Titov's avatar
Nikita Titov committed
2980
2981
2982
2983
2984

        Returns
        -------
        self : Dataset
            Dataset with set reference.
Guolin Ke's avatar
Guolin Ke committed
2985
        """
2986
2987
2988
        self.set_categorical_feature(reference.categorical_feature).set_feature_name(
            reference.feature_name
        )._set_predictor(reference._predictor)
2989
        # we're done if self and reference share a common upstream reference
2990
        if self.get_ref_chain().intersection(reference.get_ref_chain()):
Nikita Titov's avatar
Nikita Titov committed
2991
            return self
Guolin Ke's avatar
Guolin Ke committed
2992
2993
        if self.data is not None:
            self.reference = reference
Nikita Titov's avatar
Nikita Titov committed
2994
            return self._free_handle()
Guolin Ke's avatar
Guolin Ke committed
2995
        else:
2996
2997
2998
2999
            raise LightGBMError(
                "Cannot set reference after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
Guolin Ke's avatar
Guolin Ke committed
3000

3001
    def set_feature_name(self, feature_name: _LGBM_FeatureNameConfiguration) -> "Dataset":
3002
        """Set feature name.
Guolin Ke's avatar
Guolin Ke committed
3003
3004
3005

        Parameters
        ----------
3006
        feature_name : list of str
3007
            Feature names.
Nikita Titov's avatar
Nikita Titov committed
3008
3009
3010
3011
3012

        Returns
        -------
        self : Dataset
            Dataset with set feature name.
Guolin Ke's avatar
Guolin Ke committed
3013
        """
3014
        if feature_name != "auto":
3015
            self.feature_name = feature_name
3016
        if self._handle is not None and feature_name is not None and feature_name != "auto":
wxchan's avatar
wxchan committed
3017
            if len(feature_name) != self.num_feature():
3018
3019
3020
                raise ValueError(
                    f"Length of feature_name({len(feature_name)}) and num_feature({self.num_feature()}) don't match"
                )
3021
            c_feature_name = [_c_str(name) for name in feature_name]
3022
3023
3024
3025
3026
3027
3028
            _safe_call(
                _LIB.LGBM_DatasetSetFeatureNames(
                    self._handle,
                    _c_array(ctypes.c_char_p, c_feature_name),
                    ctypes.c_int(len(feature_name)),
                )
            )
Nikita Titov's avatar
Nikita Titov committed
3029
        return self
Guolin Ke's avatar
Guolin Ke committed
3030

3031
    def set_label(self, label: Optional[_LGBM_LabelType]) -> "Dataset":
3032
        """Set label of Dataset.
Guolin Ke's avatar
Guolin Ke committed
3033
3034
3035

        Parameters
        ----------
3036
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None
3037
            The label information to be set into Dataset.
Nikita Titov's avatar
Nikita Titov committed
3038
3039
3040
3041
3042

        Returns
        -------
        self : Dataset
            Dataset with set label.
Guolin Ke's avatar
Guolin Ke committed
3043
3044
        """
        self.label = label
3045
        if self._handle is not None:
3046
3047
            if isinstance(label, pd_DataFrame):
                if len(label.columns) > 1:
3048
                    raise ValueError("DataFrame for label cannot have multiple columns")
3049
                label_array = np.ravel(_pandas_to_numpy(label, target_dtype=np.float32))
3050
3051
            elif _is_pyarrow_array(label):
                label_array = label
3052
            else:
3053
3054
3055
                label_array = _list_to_1d_numpy(label, dtype=np.float32, name="label")
            self.set_field("label", label_array)
            self.label = self.get_field("label")  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
3056
        return self
Guolin Ke's avatar
Guolin Ke committed
3057

3058
3059
    def set_weight(
        self,
3060
        weight: Optional[_LGBM_WeightType],
3061
    ) -> "Dataset":
3062
        """Set weight of each instance.
Guolin Ke's avatar
Guolin Ke committed
3063
3064
3065

        Parameters
        ----------
3066
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3067
            Weight to be set for each data point. Weights should be non-negative.
Nikita Titov's avatar
Nikita Titov committed
3068
3069
3070
3071
3072

        Returns
        -------
        self : Dataset
            Dataset with set weight.
Guolin Ke's avatar
Guolin Ke committed
3073
        """
3074
3075
3076
3077
3078
3079
3080
        # Check if the weight contains values other than one
        if weight is not None:
            if _is_pyarrow_array(weight):
                if pa_compute.all(pa_compute.equal(weight, 1)).as_py():
                    weight = None
            elif np.all(weight == 1):
                weight = None
Guolin Ke's avatar
Guolin Ke committed
3081
        self.weight = weight
3082
3083

        # Set field
3084
        if self._handle is not None and weight is not None:
3085
            if not _is_pyarrow_array(weight):
3086
3087
3088
                weight = _list_to_1d_numpy(weight, dtype=np.float32, name="weight")
            self.set_field("weight", weight)
            self.weight = self.get_field("weight")  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
3089
        return self
Guolin Ke's avatar
Guolin Ke committed
3090

3091
3092
    def set_init_score(
        self,
3093
        init_score: Optional[_LGBM_InitScoreType],
3094
    ) -> "Dataset":
3095
        """Set init score of Booster to start from.
Guolin Ke's avatar
Guolin Ke committed
3096
3097
3098

        Parameters
        ----------
3099
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray, pyarrow Table (for multi-class task) or None
3100
            Init score for Booster.
Nikita Titov's avatar
Nikita Titov committed
3101
3102
3103
3104
3105

        Returns
        -------
        self : Dataset
            Dataset with set init score.
Guolin Ke's avatar
Guolin Ke committed
3106
3107
        """
        self.init_score = init_score
3108
        if self._handle is not None and init_score is not None:
3109
3110
            self.set_field("init_score", init_score)
            self.init_score = self.get_field("init_score")  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
3111
        return self
Guolin Ke's avatar
Guolin Ke committed
3112

3113
3114
    def set_group(
        self,
3115
        group: Optional[_LGBM_GroupType],
3116
    ) -> "Dataset":
3117
        """Set group size of Dataset (used for ranking).
Guolin Ke's avatar
Guolin Ke committed
3118
3119
3120

        Parameters
        ----------
3121
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3122
3123
3124
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
3125
3126
            For example, if you have a 100-document dataset with ``group = [10, 20, 40, 10, 10, 10]``, that means that you have 6 groups,
            where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.
Nikita Titov's avatar
Nikita Titov committed
3127
3128
3129
3130
3131

        Returns
        -------
        self : Dataset
            Dataset with set group.
Guolin Ke's avatar
Guolin Ke committed
3132
3133
        """
        self.group = group
3134
        if self._handle is not None and group is not None:
3135
            if not _is_pyarrow_array(group):
3136
3137
                group = _list_to_1d_numpy(group, dtype=np.int32, name="group")
            self.set_field("group", group)
3138
            # original values can be modified at cpp side
3139
            constructed_group = self.get_field("group")
3140
3141
            if constructed_group is not None:
                self.group = np.diff(constructed_group)
Nikita Titov's avatar
Nikita Titov committed
3142
        return self
Guolin Ke's avatar
Guolin Ke committed
3143

3144
3145
    def set_position(
        self,
3146
        position: Optional[_LGBM_PositionType],
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
    ) -> "Dataset":
        """Set position of Dataset (used for ranking).

        Parameters
        ----------
        position : numpy 1-D array, pandas Series or None, optional (default=None)
            Position of items used in unbiased learning-to-rank task.

        Returns
        -------
        self : Dataset
            Dataset with set position.
        """
        self.position = position
        if self._handle is not None and position is not None:
3162
3163
            position = _list_to_1d_numpy(position, dtype=np.int32, name="position")
            self.set_field("position", position)
3164
3165
        return self

3166
    def get_feature_name(self) -> List[str]:
3167
3168
3169
3170
        """Get the names of columns (features) in the Dataset.

        Returns
        -------
3171
        feature_names : list of str
3172
3173
            The names of columns (features) in the Dataset.
        """
3174
        if self._handle is None:
3175
3176
3177
3178
3179
            raise LightGBMError("Cannot get feature_name before construct dataset")
        num_feature = self.num_feature()
        tmp_out_len = ctypes.c_int(0)
        reserved_string_buffer_size = 255
        required_string_buffer_size = ctypes.c_size_t(0)
3180
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
3181
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
        _safe_call(
            _LIB.LGBM_DatasetGetFeatureNames(
                self._handle,
                ctypes.c_int(num_feature),
                ctypes.byref(tmp_out_len),
                ctypes.c_size_t(reserved_string_buffer_size),
                ctypes.byref(required_string_buffer_size),
                ptr_string_buffers,
            )
        )
3192
3193
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
3194
3195
3196
3197
        actual_string_buffer_size = required_string_buffer_size.value
        # if buffer length is not long enough, reallocate buffers
        if reserved_string_buffer_size < actual_string_buffer_size:
            string_buffers = [ctypes.create_string_buffer(actual_string_buffer_size) for _ in range(num_feature)]
3198
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
            _safe_call(
                _LIB.LGBM_DatasetGetFeatureNames(
                    self._handle,
                    ctypes.c_int(num_feature),
                    ctypes.byref(tmp_out_len),
                    ctypes.c_size_t(actual_string_buffer_size),
                    ctypes.byref(required_string_buffer_size),
                    ptr_string_buffers,
                )
            )
        return [string_buffers[i].value.decode("utf-8") for i in range(num_feature)]
3210

3211
    def get_label(self) -> Optional[_LGBM_LabelType]:
3212
        """Get the label of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3213
3214
3215

        Returns
        -------
3216
        label : list, numpy 1-D array, pandas Series / one-column DataFrame or None
3217
            The label information from the Dataset.
3218
            For a constructed ``Dataset``, this will only return a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3219
        """
3220
        if self.label is None:
3221
            self.label = self.get_field("label")
Guolin Ke's avatar
Guolin Ke committed
3222
3223
        return self.label

3224
    def get_weight(self) -> Optional[_LGBM_WeightType]:
3225
        """Get the weight of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3226
3227
3228

        Returns
        -------
3229
        weight : list, numpy 1-D array, pandas Series or None
3230
            Weight for each data point from the Dataset. Weights should be non-negative.
3231
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3232
        """
3233
        if self.weight is None:
3234
            self.weight = self.get_field("weight")
Guolin Ke's avatar
Guolin Ke committed
3235
3236
        return self.weight

3237
    def get_init_score(self) -> Optional[_LGBM_InitScoreType]:
3238
        """Get the initial score of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3239
3240
3241

        Returns
        -------
3242
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None
3243
            Init score of Booster.
3244
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3245
        """
3246
        if self.init_score is None:
3247
            self.init_score = self.get_field("init_score")
Guolin Ke's avatar
Guolin Ke committed
3248
3249
        return self.init_score

3250
    def get_data(self) -> Optional[_LGBM_TrainDataType]:
3251
3252
3253
3254
        """Get the raw data of the Dataset.

        Returns
        -------
3255
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array or None
3256
3257
            Raw data used in the Dataset construction.
        """
3258
        if self._handle is None:
3259
            raise Exception("Cannot get data before construct Dataset")
3260
        if self._need_slice and self.used_indices is not None and self.reference is not None:
Guolin Ke's avatar
Guolin Ke committed
3261
3262
            self.data = self.reference.data
            if self.data is not None:
3263
                if isinstance(self.data, (np.ndarray, scipy.sparse.spmatrix)):
Guolin Ke's avatar
Guolin Ke committed
3264
                    self.data = self.data[self.used_indices, :]
3265
                elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3266
                    self.data = self.data.iloc[self.used_indices].copy()
3267
                elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3268
                    self.data = self.data[self.used_indices, :]
3269
3270
                elif isinstance(self.data, Sequence):
                    self.data = self.data[self.used_indices]
3271
                elif _is_list_of_sequences(self.data) and len(self.data) > 0:
3272
                    self.data = np.array(list(self._yield_row_from_seqlist(self.data, self.used_indices)))
Guolin Ke's avatar
Guolin Ke committed
3273
                else:
3274
3275
3276
                    _log_warning(
                        f"Cannot subset {type(self.data).__name__} type of raw data.\n" "Returning original raw data"
                    )
3277
            self._need_slice = False
Guolin Ke's avatar
Guolin Ke committed
3278
        if self.data is None:
3279
3280
3281
3282
            raise LightGBMError(
                "Cannot call `get_data` after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
3283
3284
        return self.data

3285
    def get_group(self) -> Optional[_LGBM_GroupType]:
3286
        """Get the group of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3287
3288
3289

        Returns
        -------
3290
        group : list, numpy 1-D array, pandas Series or None
3291
3292
3293
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
3294
3295
            For example, if you have a 100-document dataset with ``group = [10, 20, 40, 10, 10, 10]``, that means that you have 6 groups,
            where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.
3296
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3297
        """
3298
        if self.group is None:
3299
            self.group = self.get_field("group")
Guolin Ke's avatar
Guolin Ke committed
3300
3301
            if self.group is not None:
                # group data from LightGBM is boundaries data, need to convert to group size
Nikita Titov's avatar
Nikita Titov committed
3302
                self.group = np.diff(self.group)
Guolin Ke's avatar
Guolin Ke committed
3303
3304
        return self.group

3305
    def get_position(self) -> Optional[_LGBM_PositionType]:
3306
3307
3308
3309
        """Get the position of the Dataset.

        Returns
        -------
3310
        position : numpy 1-D array, pandas Series or None
3311
            Position of items used in unbiased learning-to-rank task.
3312
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
3313
3314
        """
        if self.position is None:
3315
            self.position = self.get_field("position")
3316
3317
        return self.position

3318
    def num_data(self) -> int:
3319
        """Get the number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3320
3321
3322

        Returns
        -------
3323
3324
        number_of_rows : int
            The number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3325
        """
3326
        if self._handle is not None:
3327
            ret = ctypes.c_int(0)
3328
3329
3330
3331
3332
3333
            _safe_call(
                _LIB.LGBM_DatasetGetNumData(
                    self._handle,
                    ctypes.byref(ret),
                )
            )
wxchan's avatar
wxchan committed
3334
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3335
        else:
3336
            raise LightGBMError("Cannot get num_data before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3337

3338
    def num_feature(self) -> int:
3339
        """Get the number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3340
3341
3342

        Returns
        -------
3343
3344
        number_of_columns : int
            The number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3345
        """
3346
        if self._handle is not None:
3347
            ret = ctypes.c_int(0)
3348
3349
3350
3351
3352
3353
            _safe_call(
                _LIB.LGBM_DatasetGetNumFeature(
                    self._handle,
                    ctypes.byref(ret),
                )
            )
wxchan's avatar
wxchan committed
3354
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3355
        else:
3356
            raise LightGBMError("Cannot get num_feature before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3357

3358
    def feature_num_bin(self, feature: Union[int, str]) -> int:
3359
3360
        """Get the number of bins for a feature.

3361
3362
        .. versionadded:: 4.0.0

3363
3364
        Parameters
        ----------
3365
3366
        feature : int or str
            Index or name of the feature.
3367
3368
3369
3370
3371
3372

        Returns
        -------
        number_of_bins : int
            The number of constructed bins for the feature in the Dataset.
        """
3373
        if self._handle is not None:
3374
            if isinstance(feature, str):
3375
3376
3377
                feature_index = self.feature_name.index(feature)
            else:
                feature_index = feature
3378
            ret = ctypes.c_int(0)
3379
3380
3381
3382
3383
3384
3385
            _safe_call(
                _LIB.LGBM_DatasetGetFeatureNumBin(
                    self._handle,
                    ctypes.c_int(feature_index),
                    ctypes.byref(ret),
                )
            )
3386
3387
3388
3389
            return ret.value
        else:
            raise LightGBMError("Cannot get feature_num_bin before construct dataset")

3390
    def get_ref_chain(self, ref_limit: int = 100) -> Set["Dataset"]:
3391
3392
3393
3394
3395
        """Get a chain of Dataset objects.

        Starts with r, then goes to r.reference (if exists),
        then to r.reference.reference, etc.
        until we hit ``ref_limit`` or a reference loop.
3396
3397
3398
3399
3400

        Parameters
        ----------
        ref_limit : int, optional (default=100)
            The limit number of references.
3401
3402
3403

        Returns
        -------
3404
3405
3406
        ref_chain : set of Dataset
            Chain of references of the Datasets.
        """
3407
        head = self
3408
        ref_chain: Set[Dataset] = set()
3409
3410
        while len(ref_chain) < ref_limit:
            if isinstance(head, Dataset):
3411
                ref_chain.add(head)
3412
3413
3414
3415
3416
3417
                if (head.reference is not None) and (head.reference not in ref_chain):
                    head = head.reference
                else:
                    break
            else:
                break
Nikita Titov's avatar
Nikita Titov committed
3418
        return ref_chain
3419

3420
    def add_features_from(self, other: "Dataset") -> "Dataset":
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
        """Add features from other Dataset to the current Dataset.

        Both Datasets must be constructed before calling this method.

        Parameters
        ----------
        other : Dataset
            The Dataset to take features from.

        Returns
        -------
        self : Dataset
            Dataset with the new features added.
        """
3435
        if self._handle is None or other._handle is None:
3436
3437
3438
3439
3440
3441
3442
            raise ValueError("Both source and target Datasets must be constructed before adding features")
        _safe_call(
            _LIB.LGBM_DatasetAddFeaturesFrom(
                self._handle,
                other._handle,
            )
        )
Guolin Ke's avatar
Guolin Ke committed
3443
3444
3445
3446
3447
3448
3449
3450
        was_none = self.data is None
        old_self_data_type = type(self.data).__name__
        if other.data is None:
            self.data = None
        elif self.data is not None:
            if isinstance(self.data, np.ndarray):
                if isinstance(other.data, np.ndarray):
                    self.data = np.hstack((self.data, other.data))
3451
                elif isinstance(other.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3452
                    self.data = np.hstack((self.data, other.data.toarray()))
3453
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3454
                    self.data = np.hstack((self.data, other.data.values))
3455
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3456
3457
3458
                    self.data = np.hstack((self.data, other.data.to_numpy()))
                else:
                    self.data = None
3459
            elif isinstance(self.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3460
                sparse_format = self.data.getformat()
3461
                if isinstance(other.data, (np.ndarray, scipy.sparse.spmatrix)):
Guolin Ke's avatar
Guolin Ke committed
3462
                    self.data = scipy.sparse.hstack((self.data, other.data), format=sparse_format)
3463
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3464
                    self.data = scipy.sparse.hstack((self.data, other.data.values), format=sparse_format)
3465
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3466
3467
3468
                    self.data = scipy.sparse.hstack((self.data, other.data.to_numpy()), format=sparse_format)
                else:
                    self.data = None
3469
            elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3470
                if not PANDAS_INSTALLED:
3471
3472
3473
3474
3475
                    raise LightGBMError(
                        "Cannot add features to DataFrame type of raw data "
                        "without pandas installed. "
                        "Install pandas and restart your session."
                    )
Guolin Ke's avatar
Guolin Ke committed
3476
                if isinstance(other.data, np.ndarray):
3477
                    self.data = concat((self.data, pd_DataFrame(other.data)), axis=1, ignore_index=True)
3478
                elif isinstance(other.data, scipy.sparse.spmatrix):
3479
                    self.data = concat((self.data, pd_DataFrame(other.data.toarray())), axis=1, ignore_index=True)
3480
                elif isinstance(other.data, pd_DataFrame):
3481
                    self.data = concat((self.data, other.data), axis=1, ignore_index=True)
3482
                elif isinstance(other.data, dt_DataTable):
3483
                    self.data = concat((self.data, pd_DataFrame(other.data.to_numpy())), axis=1, ignore_index=True)
Guolin Ke's avatar
Guolin Ke committed
3484
3485
                else:
                    self.data = None
3486
            elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3487
                if isinstance(other.data, np.ndarray):
3488
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data)))
3489
                elif isinstance(other.data, scipy.sparse.spmatrix):
3490
3491
3492
3493
3494
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.toarray())))
                elif isinstance(other.data, pd_DataFrame):
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.values)))
                elif isinstance(other.data, dt_DataTable):
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.to_numpy())))
Guolin Ke's avatar
Guolin Ke committed
3495
3496
3497
3498
3499
                else:
                    self.data = None
            else:
                self.data = None
        if self.data is None:
3500
3501
3502
3503
3504
3505
3506
            err_msg = (
                f"Cannot add features from {type(other.data).__name__} type of raw data to "
                f"{old_self_data_type} type of raw data.\n"
            )
            err_msg += (
                "Set free_raw_data=False when construct Dataset to avoid this" if was_none else "Freeing raw data"
            )
3507
            _log_warning(err_msg)
Guolin Ke's avatar
Guolin Ke committed
3508
        self.feature_name = self.get_feature_name()
3509
3510
3511
3512
        _log_warning(
            "Reseting categorical features.\n"
            "You can set new categorical features via ``set_categorical_feature`` method"
        )
Guolin Ke's avatar
Guolin Ke committed
3513
3514
        self.categorical_feature = "auto"
        self.pandas_categorical = None
3515
3516
        return self

3517
    def _dump_text(self, filename: Union[str, Path]) -> "Dataset":
3518
3519
3520
3521
3522
3523
        """Save Dataset to a text file.

        This format cannot be loaded back in by LightGBM, but is useful for debugging purposes.

        Parameters
        ----------
3524
        filename : str or pathlib.Path
3525
3526
3527
3528
3529
3530
3531
            Name of the output file.

        Returns
        -------
        self : Dataset
            Returns self.
        """
3532
3533
3534
3535
3536
3537
        _safe_call(
            _LIB.LGBM_DatasetDumpText(
                self.construct()._handle,
                _c_str(str(filename)),
            )
        )
3538
3539
        return self

wxchan's avatar
wxchan committed
3540

3541
3542
_LGBM_CustomObjectiveFunction = Callable[
    [np.ndarray, Dataset],
3543
    Tuple[np.ndarray, np.ndarray],
3544
]
3545
3546
3547
_LGBM_CustomEvalFunction = Union[
    Callable[
        [np.ndarray, Dataset],
3548
        _LGBM_EvalFunctionResultType,
3549
3550
3551
    ],
    Callable[
        [np.ndarray, Dataset],
3552
3553
        List[_LGBM_EvalFunctionResultType],
    ],
3554
]
3555
3556


3557
class Booster:
3558
    """Booster in LightGBM."""
3559

3560
3561
3562
3563
3564
    def __init__(
        self,
        params: Optional[Dict[str, Any]] = None,
        train_set: Optional[Dataset] = None,
        model_file: Optional[Union[str, Path]] = None,
3565
        model_str: Optional[str] = None,
3566
    ):
3567
        """Initialize the Booster.
wxchan's avatar
wxchan committed
3568
3569
3570

        Parameters
        ----------
Nikita Titov's avatar
Nikita Titov committed
3571
        params : dict or None, optional (default=None)
3572
3573
3574
            Parameters for Booster.
        train_set : Dataset or None, optional (default=None)
            Training dataset.
3575
        model_file : str, pathlib.Path or None, optional (default=None)
wxchan's avatar
wxchan committed
3576
            Path to the model file.
3577
        model_str : str or None, optional (default=None)
3578
            Model will be loaded from this string.
wxchan's avatar
wxchan committed
3579
        """
3580
        self._handle = ctypes.c_void_p()
3581
        self._network = False
wxchan's avatar
wxchan committed
3582
        self.__need_reload_eval_info = True
3583
        self._train_data_name = "training"
3584
        self.__set_objective_to_none = False
wxchan's avatar
wxchan committed
3585
        self.best_iteration = -1
3586
        self.best_score: _LGBM_BoosterBestScoreType = {}
3587
        params = {} if params is None else deepcopy(params)
wxchan's avatar
wxchan committed
3588
        if train_set is not None:
3589
            # Training task
wxchan's avatar
wxchan committed
3590
            if not isinstance(train_set, Dataset):
3591
                raise TypeError(f"Training data should be Dataset instance, met {type(train_set).__name__}")
3592
3593
3594
            params = _choose_param_value(
                main_param_name="machines",
                params=params,
3595
                default_value=None,
3596
3597
3598
3599
3600
3601
3602
            )
            # if "machines" is given, assume user wants to do distributed learning, and set up network
            if params["machines"] is None:
                params.pop("machines", None)
            else:
                machines = params["machines"]
                if isinstance(machines, str):
3603
                    num_machines_from_machine_list = len(machines.split(","))
3604
3605
                elif isinstance(machines, (list, set)):
                    num_machines_from_machine_list = len(machines)
3606
                    machines = ",".join(machines)
3607
3608
3609
3610
3611
3612
                else:
                    raise ValueError("Invalid machines in params.")

                params = _choose_param_value(
                    main_param_name="num_machines",
                    params=params,
3613
                    default_value=num_machines_from_machine_list,
3614
3615
3616
3617
                )
                params = _choose_param_value(
                    main_param_name="local_listen_port",
                    params=params,
3618
                    default_value=12400,
3619
3620
3621
3622
3623
                )
                self.set_network(
                    machines=machines,
                    local_listen_port=params["local_listen_port"],
                    listen_time_out=params.get("time_out", 120),
3624
                    num_machines=params["num_machines"],
3625
                )
3626
            # construct booster object
3627
3628
3629
            train_set.construct()
            # copy the parameters from train_set
            params.update(train_set.get_params())
3630
            params_str = _param_dict_to_str(params)
3631
3632
3633
3634
3635
3636
3637
            _safe_call(
                _LIB.LGBM_BoosterCreate(
                    train_set._handle,
                    _c_str(params_str),
                    ctypes.byref(self._handle),
                )
            )
3638
            # save reference to data
wxchan's avatar
wxchan committed
3639
            self.train_set = train_set
3640
3641
            self.valid_sets: List[Dataset] = []
            self.name_valid_sets: List[str] = []
wxchan's avatar
wxchan committed
3642
            self.__num_dataset = 1
Guolin Ke's avatar
Guolin Ke committed
3643
3644
            self.__init_predictor = train_set._predictor
            if self.__init_predictor is not None:
3645
3646
3647
3648
3649
3650
                _safe_call(
                    _LIB.LGBM_BoosterMerge(
                        self._handle,
                        self.__init_predictor._handle,
                    )
                )
Guolin Ke's avatar
Guolin Ke committed
3651
            out_num_class = ctypes.c_int(0)
3652
3653
3654
3655
3656
3657
            _safe_call(
                _LIB.LGBM_BoosterGetNumClasses(
                    self._handle,
                    ctypes.byref(out_num_class),
                )
            )
wxchan's avatar
wxchan committed
3658
            self.__num_class = out_num_class.value
3659
            # buffer for inner predict
3660
            self.__inner_predict_buffer: List[Optional[np.ndarray]] = [None]
wxchan's avatar
wxchan committed
3661
3662
            self.__is_predicted_cur_iter = [False]
            self.__get_eval_info()
3663
            self.pandas_categorical = train_set.pandas_categorical
3664
            self.train_set_version = train_set.version
wxchan's avatar
wxchan committed
3665
        elif model_file is not None:
3666
            # Prediction task
Guolin Ke's avatar
Guolin Ke committed
3667
            out_num_iterations = ctypes.c_int(0)
3668
3669
3670
3671
3672
3673
3674
            _safe_call(
                _LIB.LGBM_BoosterCreateFromModelfile(
                    _c_str(str(model_file)),
                    ctypes.byref(out_num_iterations),
                    ctypes.byref(self._handle),
                )
            )
Guolin Ke's avatar
Guolin Ke committed
3675
            out_num_class = ctypes.c_int(0)
3676
3677
3678
3679
3680
3681
            _safe_call(
                _LIB.LGBM_BoosterGetNumClasses(
                    self._handle,
                    ctypes.byref(out_num_class),
                )
            )
wxchan's avatar
wxchan committed
3682
            self.__num_class = out_num_class.value
3683
            self.pandas_categorical = _load_pandas_categorical(file_name=model_file)
3684
            if params:
3685
                _log_warning("Ignoring params argument, using parameters from model file.")
3686
            params = self._get_loaded_param()
3687
        elif model_str is not None:
3688
            self.model_from_string(model_str)
wxchan's avatar
wxchan committed
3689
        else:
3690
3691
3692
            raise TypeError(
                "Need at least one training dataset or model file or model string " "to create Booster instance"
            )
3693
        self.params = params
wxchan's avatar
wxchan committed
3694

3695
    def __del__(self) -> None:
3696
        try:
3697
            if self._network:
3698
3699
3700
3701
                self.free_network()
        except AttributeError:
            pass
        try:
3702
3703
            if self._handle is not None:
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
3704
3705
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
3706

3707
    def __copy__(self) -> "Booster":
wxchan's avatar
wxchan committed
3708
3709
        return self.__deepcopy__(None)

3710
    def __deepcopy__(self, *args: Any, **kwargs: Any) -> "Booster":
3711
        model_str = self.model_to_string(num_iteration=-1)
3712
        return Booster(model_str=model_str)
wxchan's avatar
wxchan committed
3713

3714
    def __getstate__(self) -> Dict[str, Any]:
wxchan's avatar
wxchan committed
3715
        this = self.__dict__.copy()
3716
3717
3718
        handle = this["_handle"]
        this.pop("train_set", None)
        this.pop("valid_sets", None)
wxchan's avatar
wxchan committed
3719
        if handle is not None:
3720
            this["_handle"] = self.model_to_string(num_iteration=-1)
wxchan's avatar
wxchan committed
3721
3722
        return this

3723
    def __setstate__(self, state: Dict[str, Any]) -> None:
3724
        model_str = state.get("_handle", state.get("handle", None))
3725
        if model_str is not None:
wxchan's avatar
wxchan committed
3726
            handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
3727
            out_num_iterations = ctypes.c_int(0)
3728
3729
3730
3731
3732
3733
3734
3735
            _safe_call(
                _LIB.LGBM_BoosterLoadModelFromString(
                    _c_str(model_str),
                    ctypes.byref(out_num_iterations),
                    ctypes.byref(handle),
                )
            )
            state["_handle"] = handle
wxchan's avatar
wxchan committed
3736
3737
        self.__dict__.update(state)

3738
3739
3740
3741
    def _get_loaded_param(self) -> Dict[str, Any]:
        buffer_len = 1 << 20
        tmp_out_len = ctypes.c_int64(0)
        string_buffer = ctypes.create_string_buffer(buffer_len)
3742
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3743
3744
3745
3746
3747
3748
3749
3750
        _safe_call(
            _LIB.LGBM_BoosterGetLoadedParam(
                self._handle,
                ctypes.c_int64(buffer_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer,
            )
        )
3751
3752
3753
3754
        actual_len = tmp_out_len.value
        # if buffer length is not long enough, re-allocate a buffer
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
3755
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3756
3757
3758
3759
3760
3761
3762
3763
3764
            _safe_call(
                _LIB.LGBM_BoosterGetLoadedParam(
                    self._handle,
                    ctypes.c_int64(actual_len),
                    ctypes.byref(tmp_out_len),
                    ptr_string_buffer,
                )
            )
        return json.loads(string_buffer.value.decode("utf-8"))
3765

3766
    def free_dataset(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3767
3768
3769
3770
3771
3772
3773
        """Free Booster's Datasets.

        Returns
        -------
        self : Booster
            Booster without Datasets.
        """
3774
3775
        self.__dict__.pop("train_set", None)
        self.__dict__.pop("valid_sets", None)
3776
        self.__num_dataset = 0
Nikita Titov's avatar
Nikita Titov committed
3777
        return self
wxchan's avatar
wxchan committed
3778

3779
    def _free_buffer(self) -> "Booster":
3780
3781
        self.__inner_predict_buffer = []
        self.__is_predicted_cur_iter = []
Nikita Titov's avatar
Nikita Titov committed
3782
        return self
3783

3784
3785
3786
3787
3788
    def set_network(
        self,
        machines: Union[List[str], Set[str], str],
        local_listen_port: int = 12400,
        listen_time_out: int = 120,
3789
        num_machines: int = 1,
3790
    ) -> "Booster":
3791
3792
3793
3794
        """Set the network configuration.

        Parameters
        ----------
3795
        machines : list, set or str
3796
            Names of machines.
Nikita Titov's avatar
Nikita Titov committed
3797
        local_listen_port : int, optional (default=12400)
3798
            TCP listen port for local machines.
Nikita Titov's avatar
Nikita Titov committed
3799
        listen_time_out : int, optional (default=120)
3800
            Socket time-out in minutes.
Nikita Titov's avatar
Nikita Titov committed
3801
        num_machines : int, optional (default=1)
3802
            The number of machines for distributed learning application.
Nikita Titov's avatar
Nikita Titov committed
3803
3804
3805
3806
3807

        Returns
        -------
        self : Booster
            Booster with set network.
3808
        """
3809
        if isinstance(machines, (list, set)):
3810
3811
3812
3813
3814
3815
3816
3817
3818
            machines = ",".join(machines)
        _safe_call(
            _LIB.LGBM_NetworkInit(
                _c_str(machines),
                ctypes.c_int(local_listen_port),
                ctypes.c_int(listen_time_out),
                ctypes.c_int(num_machines),
            )
        )
3819
        self._network = True
Nikita Titov's avatar
Nikita Titov committed
3820
        return self
3821

3822
    def free_network(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3823
3824
3825
3826
3827
3828
3829
        """Free Booster's network.

        Returns
        -------
        self : Booster
            Booster with freed network.
        """
3830
        _safe_call(_LIB.LGBM_NetworkFree())
3831
        self._network = False
Nikita Titov's avatar
Nikita Titov committed
3832
        return self
3833

3834
    def trees_to_dataframe(self) -> pd_DataFrame:
3835
3836
        """Parse the fitted model and return in an easy-to-read pandas DataFrame.

3837
3838
3839
3840
        The returned DataFrame has the following columns.

            - ``tree_index`` : int64, which tree a node belongs to. 0-based, so a value of ``6``, for example, means "this node is in the 7th tree".
            - ``node_depth`` : int64, how far a node is from the root of the tree. The root node has a value of ``1``, its direct children are ``2``, etc.
3841
3842
3843
3844
3845
            - ``node_index`` : str, unique identifier for a node.
            - ``left_child`` : str, ``node_index`` of the child node to the left of a split. ``None`` for leaf nodes.
            - ``right_child`` : str, ``node_index`` of the child node to the right of a split. ``None`` for leaf nodes.
            - ``parent_index`` : str, ``node_index`` of this node's parent. ``None`` for the root node.
            - ``split_feature`` : str, name of the feature used for splitting. ``None`` for leaf nodes.
3846
3847
            - ``split_gain`` : float64, gain from adding this split to the tree. ``NaN`` for leaf nodes.
            - ``threshold`` : float64, value of the feature used to decide which side of the split a record will go down. ``NaN`` for leaf nodes.
3848
            - ``decision_type`` : str, logical operator describing how to compare a value to ``threshold``.
3849
3850
              For example, ``split_feature = "Column_10", threshold = 15, decision_type = "<="`` means that
              records where ``Column_10 <= 15`` follow the left side of the split, otherwise follows the right side of the split. ``None`` for leaf nodes.
3851
3852
            - ``missing_direction`` : str, split direction that missing values should go to. ``None`` for leaf nodes.
            - ``missing_type`` : str, describes what types of values are treated as missing.
3853
            - ``value`` : float64, predicted value for this leaf node, multiplied by the learning rate.
3854
            - ``weight`` : float64 or int64, sum of Hessian (second-order derivative of objective), summed over observations that fall in this node.
3855
3856
            - ``count`` : int64, number of records in the training data that fall into this node.

3857
3858
3859
3860
3861
3862
        Returns
        -------
        result : pandas DataFrame
            Returns a pandas DataFrame of the parsed model.
        """
        if not PANDAS_INSTALLED:
3863
3864
3865
3866
            raise LightGBMError(
                "This method cannot be run without pandas installed. "
                "You must install pandas and restart your session to use this method."
            )
3867
3868

        if self.num_trees() == 0:
3869
            raise LightGBMError("There are no trees in this Booster and thus nothing to parse")
3870

3871
        def _is_split_node(tree: Dict[str, Any]) -> bool:
3872
            return "split_index" in tree.keys()
3873

3874
3875
3876
3877
3878
        def create_node_record(
            tree: Dict[str, Any],
            node_depth: int = 1,
            tree_index: Optional[int] = None,
            feature_names: Optional[List[str]] = None,
3879
            parent_node: Optional[str] = None,
3880
3881
3882
        ) -> Dict[str, Any]:
            def _get_node_index(
                tree: Dict[str, Any],
3883
                tree_index: Optional[int],
3884
            ) -> str:
3885
                tree_num = f"{tree_index}-" if tree_index is not None else ""
3886
                is_split = _is_split_node(tree)
3887
                node_type = "S" if is_split else "L"
3888
                # if a single node tree it won't have `leaf_index` so return 0
3889
                node_num = tree.get("split_index" if is_split else "leaf_index", 0)
3890
                return f"{tree_num}{node_type}{node_num}"
3891

3892
3893
            def _get_split_feature(
                tree: Dict[str, Any],
3894
                feature_names: Optional[List[str]],
3895
            ) -> Optional[str]:
3896
3897
                if _is_split_node(tree):
                    if feature_names is not None:
3898
                        feature_name = feature_names[tree["split_feature"]]
3899
                    else:
3900
                        feature_name = tree["split_feature"]
3901
3902
3903
3904
                else:
                    feature_name = None
                return feature_name

3905
            def _is_single_node_tree(tree: Dict[str, Any]) -> bool:
3906
                return set(tree.keys()) == {"leaf_value"}
3907
3908

            # Create the node record, and populate universal data members
3909
            node: Dict[str, Union[int, str, None]] = OrderedDict()
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
            node["tree_index"] = tree_index
            node["node_depth"] = node_depth
            node["node_index"] = _get_node_index(tree, tree_index)
            node["left_child"] = None
            node["right_child"] = None
            node["parent_index"] = parent_node
            node["split_feature"] = _get_split_feature(tree, feature_names)
            node["split_gain"] = None
            node["threshold"] = None
            node["decision_type"] = None
            node["missing_direction"] = None
            node["missing_type"] = None
            node["value"] = None
            node["weight"] = None
            node["count"] = None
3925
3926
3927

            # Update values to reflect node type (leaf or split)
            if _is_split_node(tree):
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
                node["left_child"] = _get_node_index(tree["left_child"], tree_index)
                node["right_child"] = _get_node_index(tree["right_child"], tree_index)
                node["split_gain"] = tree["split_gain"]
                node["threshold"] = tree["threshold"]
                node["decision_type"] = tree["decision_type"]
                node["missing_direction"] = "left" if tree["default_left"] else "right"
                node["missing_type"] = tree["missing_type"]
                node["value"] = tree["internal_value"]
                node["weight"] = tree["internal_weight"]
                node["count"] = tree["internal_count"]
3938
            else:
3939
                node["value"] = tree["leaf_value"]
3940
                if not _is_single_node_tree(tree):
3941
3942
                    node["weight"] = tree["leaf_weight"]
                    node["count"] = tree["leaf_count"]
3943
3944
3945

            return node

3946
3947
3948
3949
3950
        def tree_dict_to_node_list(
            tree: Dict[str, Any],
            node_depth: int = 1,
            tree_index: Optional[int] = None,
            feature_names: Optional[List[str]] = None,
3951
            parent_node: Optional[str] = None,
3952
        ) -> List[Dict[str, Any]]:
3953
3954
3955
3956
3957
3958
3959
            node = create_node_record(
                tree=tree,
                node_depth=node_depth,
                tree_index=tree_index,
                feature_names=feature_names,
                parent_node=parent_node,
            )
3960
3961
3962
3963
3964

            res = [node]

            if _is_split_node(tree):
                # traverse the next level of the tree
3965
                children = ["left_child", "right_child"]
3966
3967
                for child in children:
                    subtree_list = tree_dict_to_node_list(
3968
                        tree=tree[child],
3969
3970
3971
                        node_depth=node_depth + 1,
                        tree_index=tree_index,
                        feature_names=feature_names,
3972
                        parent_node=node["node_index"],
3973
                    )
3974
3975
3976
3977
3978
3979
                    # In tree format, "subtree_list" is a list of node records (dicts),
                    # and we add node to the list.
                    res.extend(subtree_list)
            return res

        model_dict = self.dump_model()
3980
        feature_names = model_dict["feature_names"]
3981
        model_list = []
3982
3983
3984
3985
3986
3987
        for tree in model_dict["tree_info"]:
            model_list.extend(
                tree_dict_to_node_list(
                    tree=tree["tree_structure"], tree_index=tree["tree_index"], feature_names=feature_names
                )
            )
3988

3989
        return pd_DataFrame(model_list, columns=model_list[0].keys())
3990

3991
    def set_train_data_name(self, name: str) -> "Booster":
3992
3993
3994
3995
        """Set the name to the training Dataset.

        Parameters
        ----------
3996
        name : str
Nikita Titov's avatar
Nikita Titov committed
3997
3998
3999
4000
4001
4002
            Name for the training Dataset.

        Returns
        -------
        self : Booster
            Booster with set training Dataset name.
4003
        """
4004
        self._train_data_name = name
Nikita Titov's avatar
Nikita Titov committed
4005
        return self
wxchan's avatar
wxchan committed
4006

4007
    def add_valid(self, data: Dataset, name: str) -> "Booster":
4008
        """Add validation data.
wxchan's avatar
wxchan committed
4009
4010
4011
4012

        Parameters
        ----------
        data : Dataset
4013
            Validation data.
4014
        name : str
4015
            Name of validation data.
Nikita Titov's avatar
Nikita Titov committed
4016
4017
4018
4019
4020

        Returns
        -------
        self : Booster
            Booster with set validation data.
wxchan's avatar
wxchan committed
4021
        """
Guolin Ke's avatar
Guolin Ke committed
4022
        if not isinstance(data, Dataset):
4023
            raise TypeError(f"Validation data should be Dataset instance, met {type(data).__name__}")
Guolin Ke's avatar
Guolin Ke committed
4024
        if data._predictor is not self.__init_predictor:
4025
4026
4027
4028
4029
4030
4031
            raise LightGBMError("Add validation data failed, " "you should use same predictor for these data")
        _safe_call(
            _LIB.LGBM_BoosterAddValidData(
                self._handle,
                data.construct()._handle,
            )
        )
wxchan's avatar
wxchan committed
4032
4033
4034
4035
4036
        self.valid_sets.append(data)
        self.name_valid_sets.append(name)
        self.__num_dataset += 1
        self.__inner_predict_buffer.append(None)
        self.__is_predicted_cur_iter.append(False)
Nikita Titov's avatar
Nikita Titov committed
4037
        return self
wxchan's avatar
wxchan committed
4038

4039
    def reset_parameter(self, params: Dict[str, Any]) -> "Booster":
4040
        """Reset parameters of Booster.
wxchan's avatar
wxchan committed
4041
4042
4043
4044

        Parameters
        ----------
        params : dict
4045
            New parameters for Booster.
Nikita Titov's avatar
Nikita Titov committed
4046
4047
4048
4049
4050

        Returns
        -------
        self : Booster
            Booster with new parameters.
wxchan's avatar
wxchan committed
4051
        """
4052
        params_str = _param_dict_to_str(params)
wxchan's avatar
wxchan committed
4053
        if params_str:
4054
4055
4056
4057
4058
4059
            _safe_call(
                _LIB.LGBM_BoosterResetParameter(
                    self._handle,
                    _c_str(params_str),
                )
            )
Guolin Ke's avatar
Guolin Ke committed
4060
        self.params.update(params)
Nikita Titov's avatar
Nikita Titov committed
4061
        return self
wxchan's avatar
wxchan committed
4062

4063
4064
4065
    def update(
        self,
        train_set: Optional[Dataset] = None,
4066
        fobj: Optional[_LGBM_CustomObjectiveFunction] = None,
4067
    ) -> bool:
Nikita Titov's avatar
Nikita Titov committed
4068
        """Update Booster for one iteration.
4069

wxchan's avatar
wxchan committed
4070
4071
        Parameters
        ----------
4072
4073
4074
4075
        train_set : Dataset or None, optional (default=None)
            Training data.
            If None, last training data is used.
        fobj : callable or None, optional (default=None)
wxchan's avatar
wxchan committed
4076
            Customized objective function.
4077
4078
4079
            Should accept two parameters: preds, train_data,
            and return (grad, hess).

4080
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4081
                    The predicted values.
4082
4083
                    Predicted values are returned before any transformation,
                    e.g. they are raw margin instead of probability of positive class for binary task.
4084
4085
                train_data : Dataset
                    The training dataset.
4086
                grad : numpy 1-D array or numpy 2-D array (for multi-class task)
4087
4088
                    The value of the first order derivative (gradient) of the loss
                    with respect to the elements of preds for each sample point.
4089
                hess : numpy 1-D array or numpy 2-D array (for multi-class task)
4090
4091
                    The value of the second order derivative (Hessian) of the loss
                    with respect to the elements of preds for each sample point.
wxchan's avatar
wxchan committed
4092

4093
            For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes],
4094
            and grad and hess should be returned in the same format.
4095

wxchan's avatar
wxchan committed
4096
4097
        Returns
        -------
4098
4099
        is_finished : bool
            Whether the update was successfully finished.
wxchan's avatar
wxchan committed
4100
        """
4101
        # need reset training data
4102
4103
4104
4105
4106
4107
        if train_set is None and self.train_set_version != self.train_set.version:
            train_set = self.train_set
            is_the_same_train_set = False
        else:
            is_the_same_train_set = train_set is self.train_set and self.train_set_version == train_set.version
        if train_set is not None and not is_the_same_train_set:
Guolin Ke's avatar
Guolin Ke committed
4108
            if not isinstance(train_set, Dataset):
4109
                raise TypeError(f"Training data should be Dataset instance, met {type(train_set).__name__}")
Guolin Ke's avatar
Guolin Ke committed
4110
            if train_set._predictor is not self.__init_predictor:
4111
                raise LightGBMError("Replace training data failed, " "you should use same predictor for these data")
wxchan's avatar
wxchan committed
4112
            self.train_set = train_set
4113
4114
4115
4116
4117
4118
            _safe_call(
                _LIB.LGBM_BoosterResetTrainingData(
                    self._handle,
                    self.train_set.construct()._handle,
                )
            )
wxchan's avatar
wxchan committed
4119
            self.__inner_predict_buffer[0] = None
4120
            self.train_set_version = self.train_set.version
wxchan's avatar
wxchan committed
4121
4122
        is_finished = ctypes.c_int(0)
        if fobj is None:
4123
            if self.__set_objective_to_none:
4124
4125
4126
4127
4128
4129
4130
                raise LightGBMError("Cannot update due to null objective function.")
            _safe_call(
                _LIB.LGBM_BoosterUpdateOneIter(
                    self._handle,
                    ctypes.byref(is_finished),
                )
            )
4131
            self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
4132
4133
            return is_finished.value == 1
        else:
4134
            if not self.__set_objective_to_none:
Nikita Titov's avatar
Nikita Titov committed
4135
                self.reset_parameter({"objective": "none"}).__set_objective_to_none = True
wxchan's avatar
wxchan committed
4136
4137
4138
            grad, hess = fobj(self.__inner_predict(0), self.train_set)
            return self.__boost(grad, hess)

4139
4140
4141
    def __boost(
        self,
        grad: np.ndarray,
4142
        hess: np.ndarray,
4143
    ) -> bool:
4144
        """Boost Booster for one iteration with customized gradient statistics.
Nikita Titov's avatar
Nikita Titov committed
4145

Nikita Titov's avatar
Nikita Titov committed
4146
4147
        .. note::

4148
4149
            Score is returned before any transformation,
            e.g. it is raw margin instead of probability of positive class for binary task.
4150
            For multi-class task, score are numpy 2-D array of shape = [n_samples, n_classes],
4151
            and grad and hess should be returned in the same format.
4152

wxchan's avatar
wxchan committed
4153
4154
        Parameters
        ----------
4155
        grad : numpy 1-D array or numpy 2-D array (for multi-class task)
4156
4157
            The value of the first order derivative (gradient) of the loss
            with respect to the elements of score for each sample point.
4158
        hess : numpy 1-D array or numpy 2-D array (for multi-class task)
4159
4160
            The value of the second order derivative (Hessian) of the loss
            with respect to the elements of score for each sample point.
wxchan's avatar
wxchan committed
4161
4162
4163

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4164
4165
        is_finished : bool
            Whether the boost was successfully finished.
wxchan's avatar
wxchan committed
4166
        """
4167
        if self.__num_class > 1:
4168
4169
4170
4171
            grad = grad.ravel(order="F")
            hess = hess.ravel(order="F")
        grad = _list_to_1d_numpy(grad, dtype=np.float32, name="gradient")
        hess = _list_to_1d_numpy(hess, dtype=np.float32, name="hessian")
4172
4173
        assert grad.flags.c_contiguous
        assert hess.flags.c_contiguous
wxchan's avatar
wxchan committed
4174
        if len(grad) != len(hess):
4175
4176
            raise ValueError(f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) don't match")
        num_train_data = self.train_set.num_data()
4177
        if len(grad) != num_train_data * self.__num_class:
4178
4179
4180
            raise ValueError(
                f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) "
                f"don't match training data length ({num_train_data}) * "
4181
                f"number of models per one iteration ({self.__num_class})"
4182
            )
wxchan's avatar
wxchan committed
4183
        is_finished = ctypes.c_int(0)
4184
4185
4186
4187
4188
4189
4190
4191
        _safe_call(
            _LIB.LGBM_BoosterUpdateOneIterCustom(
                self._handle,
                grad.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
                hess.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
                ctypes.byref(is_finished),
            )
        )
4192
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
4193
4194
        return is_finished.value == 1

4195
    def rollback_one_iter(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
4196
4197
4198
4199
4200
4201
4202
        """Rollback one iteration.

        Returns
        -------
        self : Booster
            Booster with rolled back one iteration.
        """
4203
        _safe_call(_LIB.LGBM_BoosterRollbackOneIter(self._handle))
4204
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
Nikita Titov's avatar
Nikita Titov committed
4205
        return self
wxchan's avatar
wxchan committed
4206

4207
    def current_iteration(self) -> int:
4208
4209
4210
4211
4212
4213
4214
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
Guolin Ke's avatar
Guolin Ke committed
4215
        out_cur_iter = ctypes.c_int(0)
4216
4217
4218
4219
4220
4221
        _safe_call(
            _LIB.LGBM_BoosterGetCurrentIteration(
                self._handle,
                ctypes.byref(out_cur_iter),
            )
        )
wxchan's avatar
wxchan committed
4222
4223
        return out_cur_iter.value

4224
    def num_model_per_iteration(self) -> int:
4225
4226
4227
4228
4229
4230
4231
4232
        """Get number of models per iteration.

        Returns
        -------
        model_per_iter : int
            The number of models per iteration.
        """
        model_per_iter = ctypes.c_int(0)
4233
4234
4235
4236
4237
4238
        _safe_call(
            _LIB.LGBM_BoosterNumModelPerIteration(
                self._handle,
                ctypes.byref(model_per_iter),
            )
        )
4239
4240
        return model_per_iter.value

4241
    def num_trees(self) -> int:
4242
4243
4244
4245
4246
4247
4248
4249
        """Get number of weak sub-models.

        Returns
        -------
        num_trees : int
            The number of weak sub-models.
        """
        num_trees = ctypes.c_int(0)
4250
4251
4252
4253
4254
4255
        _safe_call(
            _LIB.LGBM_BoosterNumberOfTotalModel(
                self._handle,
                ctypes.byref(num_trees),
            )
        )
4256
4257
        return num_trees.value

4258
    def upper_bound(self) -> float:
4259
4260
4261
4262
        """Get upper bound value of a model.

        Returns
        -------
4263
        upper_bound : float
4264
4265
4266
            Upper bound value of the model.
        """
        ret = ctypes.c_double(0)
4267
4268
4269
4270
4271
4272
        _safe_call(
            _LIB.LGBM_BoosterGetUpperBoundValue(
                self._handle,
                ctypes.byref(ret),
            )
        )
4273
4274
        return ret.value

4275
    def lower_bound(self) -> float:
4276
4277
4278
4279
        """Get lower bound value of a model.

        Returns
        -------
4280
        lower_bound : float
4281
4282
4283
            Lower bound value of the model.
        """
        ret = ctypes.c_double(0)
4284
4285
4286
4287
4288
4289
        _safe_call(
            _LIB.LGBM_BoosterGetLowerBoundValue(
                self._handle,
                ctypes.byref(ret),
            )
        )
4290
4291
        return ret.value

4292
4293
4294
4295
    def eval(
        self,
        data: Dataset,
        name: str,
4296
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4297
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4298
        """Evaluate for data.
wxchan's avatar
wxchan committed
4299
4300
4301

        Parameters
        ----------
4302
4303
        data : Dataset
            Data for the evaluating.
4304
        name : str
4305
            Name of the data.
4306
        feval : callable, list of callable, or None, optional (default=None)
4307
            Customized evaluation function.
4308
            Each evaluation function should accept two parameters: preds, eval_data,
4309
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4310

4311
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4312
                    The predicted values.
4313
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4314
                    If custom objective function is used, predicted values are returned before any transformation,
4315
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
4316
                eval_data : Dataset
4317
                    A ``Dataset`` to evaluate.
4318
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4319
                    The name of evaluation function (without whitespace).
4320
4321
4322
4323
4324
                eval_result : float
                    The eval result.
                is_higher_better : bool
                    Is eval result higher better, e.g. AUC is ``is_higher_better``.

wxchan's avatar
wxchan committed
4325
4326
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4327
        result : list
4328
            List with (dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4329
        """
Guolin Ke's avatar
Guolin Ke committed
4330
4331
        if not isinstance(data, Dataset):
            raise TypeError("Can only eval for Dataset instance")
wxchan's avatar
wxchan committed
4332
4333
4334
4335
        data_idx = -1
        if data is self.train_set:
            data_idx = 0
        else:
4336
            for i in range(len(self.valid_sets)):
wxchan's avatar
wxchan committed
4337
4338
4339
                if data is self.valid_sets[i]:
                    data_idx = i + 1
                    break
4340
        # need to push new valid data
wxchan's avatar
wxchan committed
4341
4342
4343
4344
4345
4346
        if data_idx == -1:
            self.add_valid(data, name)
            data_idx = self.__num_dataset - 1

        return self.__inner_eval(name, data_idx, feval)

4347
4348
    def eval_train(
        self,
4349
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4350
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4351
        """Evaluate for training data.
wxchan's avatar
wxchan committed
4352
4353
4354

        Parameters
        ----------
4355
        feval : callable, list of callable, or None, optional (default=None)
4356
            Customized evaluation function.
4357
            Each evaluation function should accept two parameters: preds, eval_data,
4358
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4359

4360
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4361
                    The predicted values.
4362
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4363
                    If custom objective function is used, predicted values are returned before any transformation,
4364
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
Akshita Dixit's avatar
Akshita Dixit committed
4365
                eval_data : Dataset
4366
                    The training dataset.
4367
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4368
                    The name of evaluation function (without whitespace).
4369
4370
4371
4372
4373
                eval_result : float
                    The eval result.
                is_higher_better : bool
                    Is eval result higher better, e.g. AUC is ``is_higher_better``.

wxchan's avatar
wxchan committed
4374
4375
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4376
        result : list
4377
            List with (train_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4378
        """
4379
        return self.__inner_eval(self._train_data_name, 0, feval)
wxchan's avatar
wxchan committed
4380

4381
4382
    def eval_valid(
        self,
4383
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4384
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4385
        """Evaluate for validation data.
wxchan's avatar
wxchan committed
4386
4387
4388

        Parameters
        ----------
4389
        feval : callable, list of callable, or None, optional (default=None)
4390
            Customized evaluation function.
4391
            Each evaluation function should accept two parameters: preds, eval_data,
4392
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4393

4394
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4395
                    The predicted values.
4396
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4397
                    If custom objective function is used, predicted values are returned before any transformation,
4398
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
Akshita Dixit's avatar
Akshita Dixit committed
4399
                eval_data : Dataset
4400
                    The validation dataset.
4401
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4402
                    The name of evaluation function (without whitespace).
4403
4404
4405
4406
4407
                eval_result : float
                    The eval result.
                is_higher_better : bool
                    Is eval result higher better, e.g. AUC is ``is_higher_better``.

wxchan's avatar
wxchan committed
4408
4409
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4410
        result : list
4411
            List with (validation_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4412
        """
4413
4414
4415
4416
4417
        return [
            item
            for i in range(1, self.__num_dataset)
            for item in self.__inner_eval(self.name_valid_sets[i - 1], i, feval)
        ]
wxchan's avatar
wxchan committed
4418

4419
4420
4421
4422
4423
    def save_model(
        self,
        filename: Union[str, Path],
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4424
        importance_type: str = "split",
4425
    ) -> "Booster":
4426
        """Save Booster to file.
wxchan's avatar
wxchan committed
4427
4428
4429

        Parameters
        ----------
4430
        filename : str or pathlib.Path
4431
            Filename to save Booster.
4432
4433
4434
4435
        num_iteration : int or None, optional (default=None)
            Index of the iteration that should be saved.
            If None, if the best iteration exists, it is saved; otherwise, all iterations are saved.
            If <= 0, all iterations are saved.
Nikita Titov's avatar
Nikita Titov committed
4436
        start_iteration : int, optional (default=0)
4437
            Start index of the iteration that should be saved.
4438
        importance_type : str, optional (default="split")
4439
4440
4441
            What type of feature importance should be saved.
            If "split", result contains numbers of times the feature is used in a model.
            If "gain", result contains total gains of splits which use the feature.
Nikita Titov's avatar
Nikita Titov committed
4442
4443
4444
4445
4446

        Returns
        -------
        self : Booster
            Returns self.
wxchan's avatar
wxchan committed
4447
        """
4448
        if num_iteration is None:
4449
            num_iteration = self.best_iteration
4450
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4451
4452
4453
4454
4455
4456
4457
4458
4459
        _safe_call(
            _LIB.LGBM_BoosterSaveModel(
                self._handle,
                ctypes.c_int(start_iteration),
                ctypes.c_int(num_iteration),
                ctypes.c_int(importance_type_int),
                _c_str(str(filename)),
            )
        )
4460
        _dump_pandas_categorical(self.pandas_categorical, filename)
Nikita Titov's avatar
Nikita Titov committed
4461
        return self
wxchan's avatar
wxchan committed
4462

4463
4464
4465
    def shuffle_models(
        self,
        start_iteration: int = 0,
4466
        end_iteration: int = -1,
4467
    ) -> "Booster":
4468
        """Shuffle models.
Nikita Titov's avatar
Nikita Titov committed
4469

4470
4471
4472
        Parameters
        ----------
        start_iteration : int, optional (default=0)
4473
            The first iteration that will be shuffled.
4474
4475
        end_iteration : int, optional (default=-1)
            The last iteration that will be shuffled.
4476
            If <= 0, means the last available iteration.
4477

Nikita Titov's avatar
Nikita Titov committed
4478
4479
4480
4481
        Returns
        -------
        self : Booster
            Booster with shuffled models.
4482
        """
4483
4484
4485
4486
4487
4488
4489
        _safe_call(
            _LIB.LGBM_BoosterShuffleModels(
                self._handle,
                ctypes.c_int(start_iteration),
                ctypes.c_int(end_iteration),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
4490
        return self
4491

4492
    def model_from_string(self, model_str: str) -> "Booster":
4493
4494
4495
4496
        """Load Booster from a string.

        Parameters
        ----------
4497
        model_str : str
4498
4499
4500
4501
            Model will be loaded from this string.

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4502
        self : Booster
4503
4504
            Loaded Booster object.
        """
4505
4506
4507
        # ensure that existing Booster is freed before replacing it
        # with a new one createdfrom file
        _safe_call(_LIB.LGBM_BoosterFree(self._handle))
4508
        self._free_buffer()
4509
        self._handle = ctypes.c_void_p()
4510
        out_num_iterations = ctypes.c_int(0)
4511
4512
4513
4514
4515
4516
4517
        _safe_call(
            _LIB.LGBM_BoosterLoadModelFromString(
                _c_str(model_str),
                ctypes.byref(out_num_iterations),
                ctypes.byref(self._handle),
            )
        )
4518
        out_num_class = ctypes.c_int(0)
4519
4520
4521
4522
4523
4524
        _safe_call(
            _LIB.LGBM_BoosterGetNumClasses(
                self._handle,
                ctypes.byref(out_num_class),
            )
        )
4525
        self.__num_class = out_num_class.value
4526
        self.pandas_categorical = _load_pandas_categorical(model_str=model_str)
4527
4528
        return self

4529
4530
4531
4532
    def model_to_string(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4533
        importance_type: str = "split",
4534
    ) -> str:
4535
        """Save Booster to string.
4536

4537
4538
4539
4540
4541
4542
        Parameters
        ----------
        num_iteration : int or None, optional (default=None)
            Index of the iteration that should be saved.
            If None, if the best iteration exists, it is saved; otherwise, all iterations are saved.
            If <= 0, all iterations are saved.
Nikita Titov's avatar
Nikita Titov committed
4543
        start_iteration : int, optional (default=0)
4544
            Start index of the iteration that should be saved.
4545
        importance_type : str, optional (default="split")
4546
4547
4548
            What type of feature importance should be saved.
            If "split", result contains numbers of times the feature is used in a model.
            If "gain", result contains total gains of splits which use the feature.
4549
4550
4551

        Returns
        -------
4552
        str_repr : str
4553
4554
            String representation of Booster.
        """
4555
        if num_iteration is None:
4556
            num_iteration = self.best_iteration
4557
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4558
        buffer_len = 1 << 20
4559
        tmp_out_len = ctypes.c_int64(0)
4560
        string_buffer = ctypes.create_string_buffer(buffer_len)
4561
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4562
4563
        _safe_call(
            _LIB.LGBM_BoosterSaveModelToString(
4564
                self._handle,
4565
                ctypes.c_int(start_iteration),
4566
                ctypes.c_int(num_iteration),
4567
                ctypes.c_int(importance_type_int),
4568
                ctypes.c_int64(buffer_len),
4569
                ctypes.byref(tmp_out_len),
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
                ptr_string_buffer,
            )
        )
        actual_len = tmp_out_len.value
        # if buffer length is not long enough, re-allocate a buffer
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
            _safe_call(
                _LIB.LGBM_BoosterSaveModelToString(
                    self._handle,
                    ctypes.c_int(start_iteration),
                    ctypes.c_int(num_iteration),
                    ctypes.c_int(importance_type_int),
                    ctypes.c_int64(actual_len),
                    ctypes.byref(tmp_out_len),
                    ptr_string_buffer,
                )
            )
        ret = string_buffer.value.decode("utf-8")
4590
4591
        ret += _dump_pandas_categorical(self.pandas_categorical)
        return ret
4592

4593
4594
4595
4596
    def dump_model(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4597
4598
        importance_type: str = "split",
        object_hook: Optional[Callable[[Dict[str, Any]], Dict[str, Any]]] = None,
4599
    ) -> Dict[str, Any]:
Nikita Titov's avatar
Nikita Titov committed
4600
        """Dump Booster to JSON format.
wxchan's avatar
wxchan committed
4601

4602
4603
        Parameters
        ----------
4604
4605
4606
4607
        num_iteration : int or None, optional (default=None)
            Index of the iteration that should be dumped.
            If None, if the best iteration exists, it is dumped; otherwise, all iterations are dumped.
            If <= 0, all iterations are dumped.
Nikita Titov's avatar
Nikita Titov committed
4608
        start_iteration : int, optional (default=0)
4609
            Start index of the iteration that should be dumped.
4610
        importance_type : str, optional (default="split")
4611
4612
4613
            What type of feature importance should be dumped.
            If "split", result contains numbers of times the feature is used in a model.
            If "gain", result contains total gains of splits which use the feature.
4614
4615
4616
4617
4618
4619
4620
4621
4622
        object_hook : callable or None, optional (default=None)
            If not None, ``object_hook`` is a function called while parsing the json
            string returned by the C API. It may be used to alter the json, to store
            specific values while building the json structure. It avoids
            walking through the structure again. It saves a significant amount
            of time if the number of trees is huge.
            Signature is ``def object_hook(node: dict) -> dict``.
            None is equivalent to ``lambda node: node``.
            See documentation of ``json.loads()`` for further details.
4623

wxchan's avatar
wxchan committed
4624
4625
        Returns
        -------
4626
        json_repr : dict
Nikita Titov's avatar
Nikita Titov committed
4627
            JSON format of Booster.
wxchan's avatar
wxchan committed
4628
        """
4629
        if num_iteration is None:
4630
            num_iteration = self.best_iteration
4631
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
4632
        buffer_len = 1 << 20
4633
        tmp_out_len = ctypes.c_int64(0)
wxchan's avatar
wxchan committed
4634
        string_buffer = ctypes.create_string_buffer(buffer_len)
4635
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4636
4637
        _safe_call(
            _LIB.LGBM_BoosterDumpModel(
4638
                self._handle,
4639
                ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
4640
                ctypes.c_int(num_iteration),
4641
                ctypes.c_int(importance_type_int),
4642
                ctypes.c_int64(buffer_len),
wxchan's avatar
wxchan committed
4643
                ctypes.byref(tmp_out_len),
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
                ptr_string_buffer,
            )
        )
        actual_len = tmp_out_len.value
        # if buffer length is not long enough, reallocate a buffer
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
            _safe_call(
                _LIB.LGBM_BoosterDumpModel(
                    self._handle,
                    ctypes.c_int(start_iteration),
                    ctypes.c_int(num_iteration),
                    ctypes.c_int(importance_type_int),
                    ctypes.c_int64(actual_len),
                    ctypes.byref(tmp_out_len),
                    ptr_string_buffer,
                )
            )
        ret = json.loads(string_buffer.value.decode("utf-8"), object_hook=object_hook)
        ret["pandas_categorical"] = json.loads(
            json.dumps(
                self.pandas_categorical,
                default=_json_default_with_numpy,
            )
        )
4670
        return ret
wxchan's avatar
wxchan committed
4671

4672
4673
    def predict(
        self,
4674
        data: _LGBM_PredictDataType,
4675
4676
4677
4678
4679
4680
4681
        start_iteration: int = 0,
        num_iteration: Optional[int] = None,
        raw_score: bool = False,
        pred_leaf: bool = False,
        pred_contrib: bool = False,
        data_has_header: bool = False,
        validate_features: bool = False,
4682
        **kwargs: Any,
4683
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
4684
        """Make a prediction.
wxchan's avatar
wxchan committed
4685
4686
4687

        Parameters
        ----------
4688
        data : str, pathlib.Path, numpy array, pandas DataFrame, pyarrow Table, H2O DataTable's Frame or scipy.sparse
4689
            Data source for prediction.
4690
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4691
        start_iteration : int, optional (default=0)
4692
            Start index of the iteration to predict.
4693
            If <= 0, starts from the first iteration.
4694
        num_iteration : int or None, optional (default=None)
4695
4696
4697
4698
            Total number of iterations used in the prediction.
            If None, if the best iteration exists and start_iteration <= 0, the best iteration is used;
            otherwise, all iterations from ``start_iteration`` are used (no limits).
            If <= 0, all iterations from ``start_iteration`` are used (no limits).
4699
4700
4701
4702
        raw_score : bool, optional (default=False)
            Whether to predict raw scores.
        pred_leaf : bool, optional (default=False)
            Whether to predict leaf index.
4703
4704
        pred_contrib : bool, optional (default=False)
            Whether to predict feature contributions.
4705

Nikita Titov's avatar
Nikita Titov committed
4706
4707
4708
4709
4710
4711
4712
            .. note::

                If you want to get more explanations for your model's predictions using SHAP values,
                like SHAP interaction values,
                you can install the shap package (https://github.com/slundberg/shap).
                Note that unlike the shap package, with ``pred_contrib`` we return a matrix with an extra
                column, where the last column is the expected value.
4713

4714
4715
        data_has_header : bool, optional (default=False)
            Whether the data has header.
4716
            Used only if data is str.
4717
4718
4719
        validate_features : bool, optional (default=False)
            If True, ensure that the features used to predict match the ones used to train.
            Used only if data is pandas DataFrame.
4720
4721
        **kwargs
            Other parameters for the prediction.
wxchan's avatar
wxchan committed
4722
4723
4724

        Returns
        -------
4725
        result : numpy array, scipy.sparse or list of scipy.sparse
4726
            Prediction result.
4727
            Can be sparse or a list of sparse objects (each element represents predictions for one class) for feature contributions (when ``pred_contrib=True``).
wxchan's avatar
wxchan committed
4728
        """
4729
4730
4731
4732
        predictor = _InnerPredictor.from_booster(
            booster=self,
            pred_parameter=deepcopy(kwargs),
        )
4733
        if num_iteration is None:
4734
            if start_iteration <= 0:
4735
4736
4737
                num_iteration = self.best_iteration
            else:
                num_iteration = -1
4738
4739
4740
4741
4742
4743
4744
4745
        return predictor.predict(
            data=data,
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            raw_score=raw_score,
            pred_leaf=pred_leaf,
            pred_contrib=pred_contrib,
            data_has_header=data_has_header,
4746
            validate_features=validate_features,
4747
        )
wxchan's avatar
wxchan committed
4748

4749
4750
    def refit(
        self,
4751
        data: _LGBM_TrainDataType,
4752
        label: _LGBM_LabelType,
4753
4754
        decay_rate: float = 0.9,
        reference: Optional[Dataset] = None,
4755
4756
4757
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
4758
4759
        feature_name: _LGBM_FeatureNameConfiguration = "auto",
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = "auto",
4760
4761
4762
        dataset_params: Optional[Dict[str, Any]] = None,
        free_raw_data: bool = True,
        validate_features: bool = False,
4763
        **kwargs: Any,
4764
    ) -> "Booster":
Guolin Ke's avatar
Guolin Ke committed
4765
4766
4767
4768
        """Refit the existing Booster by new data.

        Parameters
        ----------
4769
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array
Guolin Ke's avatar
Guolin Ke committed
4770
            Data source for refit.
4771
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4772
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array or pyarrow ChunkedArray
Guolin Ke's avatar
Guolin Ke committed
4773
4774
            Label for refit.
        decay_rate : float, optional (default=0.9)
4775
4776
            Decay rate of refit,
            will use ``leaf_output = decay_rate * old_leaf_output + (1.0 - decay_rate) * new_leaf_output`` to refit trees.
4777
4778
        reference : Dataset or None, optional (default=None)
            Reference for ``data``.
4779
4780
4781

            .. versionadded:: 4.0.0

4782
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4783
            Weight for each ``data`` instance. Weights should be non-negative.
4784
4785
4786

            .. versionadded:: 4.0.0

4787
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4788
4789
4790
4791
4792
            Group/query size for ``data``.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
            For example, if you have a 100-document dataset with ``group = [10, 20, 40, 10, 10, 10]``, that means that you have 6 groups,
            where the first 10 records are in the first group, records 11-30 are in the second group, records 31-70 are in the third group, etc.
4793
4794
4795

            .. versionadded:: 4.0.0

4796
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray, pyarrow Table (for multi-class task) or None, optional (default=None)
4797
            Init score for ``data``.
4798
4799
4800

            .. versionadded:: 4.0.0

4801
4802
4803
        feature_name : list of str, or 'auto', optional (default="auto")
            Feature names for ``data``.
            If 'auto' and data is pandas DataFrame, data columns names are used.
4804
4805
4806

            .. versionadded:: 4.0.0

4807
4808
4809
4810
4811
        categorical_feature : list of str or int, or 'auto', optional (default="auto")
            Categorical features for ``data``.
            If list of int, interpreted as indices.
            If list of str, interpreted as feature names (need to specify ``feature_name`` as well).
            If 'auto' and data is pandas DataFrame, pandas unordered categorical columns are used.
4812
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
4813
4814
4815
            Large values could be memory consuming. Consider using consecutive integers starting from zero.
            All negative values in categorical features will be treated as missing values.
            The output cannot be monotonically constrained with respect to a categorical feature.
4816
            Floating point numbers in categorical features will be rounded towards 0.
4817
4818
4819

            .. versionadded:: 4.0.0

4820
4821
        dataset_params : dict or None, optional (default=None)
            Other parameters for Dataset ``data``.
4822
4823
4824

            .. versionadded:: 4.0.0

4825
4826
        free_raw_data : bool, optional (default=True)
            If True, raw data is freed after constructing inner Dataset for ``data``.
4827
4828
4829

            .. versionadded:: 4.0.0

4830
4831
4832
        validate_features : bool, optional (default=False)
            If True, ensure that the features used to refit the model match the original ones.
            Used only if data is pandas DataFrame.
4833
4834
4835

            .. versionadded:: 4.0.0

4836
4837
        **kwargs
            Other parameters for refit.
4838
            These parameters will be passed to ``predict`` method.
Guolin Ke's avatar
Guolin Ke committed
4839
4840
4841
4842
4843
4844

        Returns
        -------
        result : Booster
            Refitted Booster.
        """
4845
        if self.__set_objective_to_none:
4846
            raise LightGBMError("Cannot refit due to null objective function.")
4847
4848
        if dataset_params is None:
            dataset_params = {}
4849
        predictor = _InnerPredictor.from_booster(booster=self, pred_parameter=deepcopy(kwargs))
4850
        leaf_preds: np.ndarray = predictor.predict(  # type: ignore[assignment]
4851
4852
4853
            data=data,
            start_iteration=-1,
            pred_leaf=True,
4854
            validate_features=validate_features,
4855
        )
4856
        nrow, ncol = leaf_preds.shape
4857
        out_is_linear = ctypes.c_int(0)
4858
4859
4860
4861
4862
4863
        _safe_call(
            _LIB.LGBM_BoosterGetLinear(
                self._handle,
                ctypes.byref(out_is_linear),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
4864
4865
4866
        new_params = _choose_param_value(
            main_param_name="linear_tree",
            params=self.params,
4867
            default_value=None,
Nikita Titov's avatar
Nikita Titov committed
4868
        )
4869
        new_params["linear_tree"] = bool(out_is_linear.value)
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
        new_params.update(dataset_params)
        train_set = Dataset(
            data=data,
            label=label,
            reference=reference,
            weight=weight,
            group=group,
            init_score=init_score,
            feature_name=feature_name,
            categorical_feature=categorical_feature,
            params=new_params,
            free_raw_data=free_raw_data,
        )
4883
        new_params["refit_decay_rate"] = decay_rate
4884
        new_booster = Booster(new_params, train_set)
Guolin Ke's avatar
Guolin Ke committed
4885
        # Copy models
4886
4887
4888
4889
4890
4891
        _safe_call(
            _LIB.LGBM_BoosterMerge(
                new_booster._handle,
                predictor._handle,
            )
        )
Guolin Ke's avatar
Guolin Ke committed
4892
        leaf_preds = leaf_preds.reshape(-1)
4893
        ptr_data, _, _ = _c_int_array(leaf_preds)
4894
4895
4896
4897
4898
4899
4900
4901
        _safe_call(
            _LIB.LGBM_BoosterRefit(
                new_booster._handle,
                ptr_data,
                ctypes.c_int32(nrow),
                ctypes.c_int32(ncol),
            )
        )
4902
        new_booster._network = self._network
Guolin Ke's avatar
Guolin Ke committed
4903
4904
        return new_booster

4905
    def get_leaf_output(self, tree_id: int, leaf_id: int) -> float:
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
        """Get the output of a leaf.

        Parameters
        ----------
        tree_id : int
            The index of the tree.
        leaf_id : int
            The index of the leaf in the tree.

        Returns
        -------
        result : float
            The output of the leaf.
        """
4920
        ret = ctypes.c_double(0)
4921
4922
4923
4924
4925
4926
4927
4928
        _safe_call(
            _LIB.LGBM_BoosterGetLeafValue(
                self._handle,
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
                ctypes.byref(ret),
            )
        )
4929
4930
        return ret.value

4931
4932
4933
4934
4935
    def set_leaf_output(
        self,
        tree_id: int,
        leaf_id: int,
        value: float,
4936
    ) -> "Booster":
4937
4938
        """Set the output of a leaf.

4939
4940
        .. versionadded:: 4.0.0

4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
        Parameters
        ----------
        tree_id : int
            The index of the tree.
        leaf_id : int
            The index of the leaf in the tree.
        value : float
            Value to set as the output of the leaf.

        Returns
        -------
        self : Booster
            Booster with the leaf output set.
        """
        _safe_call(
            _LIB.LGBM_BoosterSetLeafValue(
4957
                self._handle,
4958
4959
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
4960
                ctypes.c_double(value),
4961
4962
4963
4964
            )
        )
        return self

4965
    def num_feature(self) -> int:
4966
4967
4968
4969
4970
4971
4972
        """Get number of features.

        Returns
        -------
        num_feature : int
            The number of features.
        """
4973
        out_num_feature = ctypes.c_int(0)
4974
4975
4976
4977
4978
4979
        _safe_call(
            _LIB.LGBM_BoosterGetNumFeature(
                self._handle,
                ctypes.byref(out_num_feature),
            )
        )
4980
4981
        return out_num_feature.value

4982
    def feature_name(self) -> List[str]:
4983
        """Get names of features.
wxchan's avatar
wxchan committed
4984
4985
4986

        Returns
        -------
4987
        result : list of str
4988
            List with names of features.
wxchan's avatar
wxchan committed
4989
        """
4990
        num_feature = self.num_feature()
4991
        # Get name of features
wxchan's avatar
wxchan committed
4992
        tmp_out_len = ctypes.c_int(0)
4993
4994
        reserved_string_buffer_size = 255
        required_string_buffer_size = ctypes.c_size_t(0)
4995
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
4996
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
        _safe_call(
            _LIB.LGBM_BoosterGetFeatureNames(
                self._handle,
                ctypes.c_int(num_feature),
                ctypes.byref(tmp_out_len),
                ctypes.c_size_t(reserved_string_buffer_size),
                ctypes.byref(required_string_buffer_size),
                ptr_string_buffers,
            )
        )
wxchan's avatar
wxchan committed
5007
5008
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
5009
5010
5011
5012
        actual_string_buffer_size = required_string_buffer_size.value
        # if buffer length is not long enough, reallocate buffers
        if reserved_string_buffer_size < actual_string_buffer_size:
            string_buffers = [ctypes.create_string_buffer(actual_string_buffer_size) for _ in range(num_feature)]
5013
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
            _safe_call(
                _LIB.LGBM_BoosterGetFeatureNames(
                    self._handle,
                    ctypes.c_int(num_feature),
                    ctypes.byref(tmp_out_len),
                    ctypes.c_size_t(actual_string_buffer_size),
                    ctypes.byref(required_string_buffer_size),
                    ptr_string_buffers,
                )
            )
        return [string_buffers[i].value.decode("utf-8") for i in range(num_feature)]
wxchan's avatar
wxchan committed
5025

5026
5027
    def feature_importance(
        self,
5028
5029
        importance_type: str = "split",
        iteration: Optional[int] = None,
5030
    ) -> np.ndarray:
5031
        """Get feature importances.
5032

5033
5034
        Parameters
        ----------
5035
        importance_type : str, optional (default="split")
5036
5037
5038
            How the importance is calculated.
            If "split", result contains numbers of times the feature is used in a model.
            If "gain", result contains total gains of splits which use the feature.
5039
5040
5041
5042
        iteration : int or None, optional (default=None)
            Limit number of iterations in the feature importance calculation.
            If None, if the best iteration exists, it is used; otherwise, all trees are used.
            If <= 0, all trees are used (no limits).
5043

5044
5045
        Returns
        -------
5046
5047
        result : numpy array
            Array with feature importances.
5048
        """
5049
5050
        if iteration is None:
            iteration = self.best_iteration
5051
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
5052
        result = np.empty(self.num_feature(), dtype=np.float64)
5053
5054
5055
5056
5057
5058
5059
5060
        _safe_call(
            _LIB.LGBM_BoosterFeatureImportance(
                self._handle,
                ctypes.c_int(iteration),
                ctypes.c_int(importance_type_int),
                result.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
            )
        )
5061
        if importance_type_int == _C_API_FEATURE_IMPORTANCE_SPLIT:
5062
            return result.astype(np.int32)
5063
5064
        else:
            return result
5065

5066
5067
5068
5069
    def get_split_value_histogram(
        self,
        feature: Union[int, str],
        bins: Optional[Union[int, str]] = None,
5070
        xgboost_style: bool = False,
5071
    ) -> Union[Tuple[np.ndarray, np.ndarray], np.ndarray, pd_DataFrame]:
5072
5073
5074
5075
        """Get split value histogram for the specified feature.

        Parameters
        ----------
5076
        feature : int or str
5077
5078
            The feature name or index the histogram is calculated for.
            If int, interpreted as index.
5079
            If str, interpreted as name.
5080

Nikita Titov's avatar
Nikita Titov committed
5081
5082
5083
            .. warning::

                Categorical features are not supported.
5084

5085
        bins : int, str or None, optional (default=None)
5086
5087
5088
            The maximum number of bins.
            If None, or int and > number of unique split values and ``xgboost_style=True``,
            the number of bins equals number of unique split values.
5089
            If str, it should be one from the list of the supported values by ``numpy.histogram()`` function.
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
        xgboost_style : bool, optional (default=False)
            Whether the returned result should be in the same form as it is in XGBoost.
            If False, the returned value is tuple of 2 numpy arrays as it is in ``numpy.histogram()`` function.
            If True, the returned value is matrix, in which the first column is the right edges of non-empty bins
            and the second one is the histogram values.

        Returns
        -------
        result_tuple : tuple of 2 numpy arrays
            If ``xgboost_style=False``, the values of the histogram of used splitting values for the specified feature
            and the bin edges.
        result_array_like : numpy array or pandas DataFrame (if pandas is installed)
            If ``xgboost_style=True``, the histogram of used splitting values for the specified feature.
        """
5104

5105
        def add(root: Dict[str, Any]) -> None:
5106
            """Recursively add thresholds."""
5107
            if "split_index" in root:  # non-leaf
5108
                if feature_names is not None and isinstance(feature, str):
5109
                    split_feature = feature_names[root["split_feature"]]
5110
                else:
5111
                    split_feature = root["split_feature"]
5112
                if split_feature == feature:
5113
5114
                    if isinstance(root["threshold"], str):
                        raise LightGBMError("Cannot compute split value histogram for the categorical feature")
5115
                    else:
5116
5117
5118
                        values.append(root["threshold"])
                add(root["left_child"])
                add(root["right_child"])
5119
5120

        model = self.dump_model()
5121
5122
        feature_names = model.get("feature_names")
        tree_infos = model["tree_info"]
5123
        values: List[float] = []
5124
        for tree_info in tree_infos:
5125
            add(tree_info["tree_structure"])
5126

5127
        if bins is None or isinstance(bins, int) and xgboost_style:
5128
5129
5130
5131
5132
5133
5134
            n_unique = len(np.unique(values))
            bins = max(min(n_unique, bins) if bins is not None else n_unique, 1)
        hist, bin_edges = np.histogram(values, bins=bins)
        if xgboost_style:
            ret = np.column_stack((bin_edges[1:], hist))
            ret = ret[ret[:, 1] > 0]
            if PANDAS_INSTALLED:
5135
                return pd_DataFrame(ret, columns=["SplitValue", "Count"])
5136
5137
5138
5139
5140
            else:
                return ret
        else:
            return hist, bin_edges

5141
5142
5143
5144
    def __inner_eval(
        self,
        data_name: str,
        data_idx: int,
5145
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]],
5146
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
5147
        """Evaluate training or validation data."""
wxchan's avatar
wxchan committed
5148
        if data_idx >= self.__num_dataset:
5149
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
5150
5151
5152
        self.__get_eval_info()
        ret = []
        if self.__num_inner_eval > 0:
5153
            result = np.empty(self.__num_inner_eval, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
5154
            tmp_out_len = ctypes.c_int(0)
5155
5156
5157
5158
5159
5160
5161
5162
            _safe_call(
                _LIB.LGBM_BoosterGetEval(
                    self._handle,
                    ctypes.c_int(data_idx),
                    ctypes.byref(tmp_out_len),
                    result.ctypes.data_as(ctypes.POINTER(ctypes.c_double)),
                )
            )
wxchan's avatar
wxchan committed
5163
            if tmp_out_len.value != self.__num_inner_eval:
5164
                raise ValueError("Wrong length of eval results")
5165
            for i in range(self.__num_inner_eval):
5166
                ret.append((data_name, self.__name_inner_eval[i], result[i], self.__higher_better_inner_eval[i]))
5167
5168
        if callable(feval):
            feval = [feval]
wxchan's avatar
wxchan committed
5169
5170
5171
5172
5173
        if feval is not None:
            if data_idx == 0:
                cur_data = self.train_set
            else:
                cur_data = self.valid_sets[data_idx - 1]
5174
5175
5176
5177
5178
5179
5180
5181
5182
            for eval_function in feval:
                if eval_function is None:
                    continue
                feval_ret = eval_function(self.__inner_predict(data_idx), cur_data)
                if isinstance(feval_ret, list):
                    for eval_name, val, is_higher_better in feval_ret:
                        ret.append((data_name, eval_name, val, is_higher_better))
                else:
                    eval_name, val, is_higher_better = feval_ret
wxchan's avatar
wxchan committed
5183
5184
5185
                    ret.append((data_name, eval_name, val, is_higher_better))
        return ret

5186
    def __inner_predict(self, data_idx: int) -> np.ndarray:
5187
        """Predict for training and validation dataset."""
wxchan's avatar
wxchan committed
5188
        if data_idx >= self.__num_dataset:
5189
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
5190
5191
5192
5193
5194
        if self.__inner_predict_buffer[data_idx] is None:
            if data_idx == 0:
                n_preds = self.train_set.num_data() * self.__num_class
            else:
                n_preds = self.valid_sets[data_idx - 1].num_data() * self.__num_class
5195
            self.__inner_predict_buffer[data_idx] = np.empty(n_preds, dtype=np.float64)
5196
        # avoid to predict many time in one iteration
wxchan's avatar
wxchan committed
5197
5198
        if not self.__is_predicted_cur_iter[data_idx]:
            tmp_out_len = ctypes.c_int64(0)
5199
            data_ptr = self.__inner_predict_buffer[data_idx].ctypes.data_as(ctypes.POINTER(ctypes.c_double))  # type: ignore[union-attr]
5200
5201
5202
5203
5204
5205
5206
5207
            _safe_call(
                _LIB.LGBM_BoosterGetPredict(
                    self._handle,
                    ctypes.c_int(data_idx),
                    ctypes.byref(tmp_out_len),
                    data_ptr,
                )
            )
5208
            if tmp_out_len.value != len(self.__inner_predict_buffer[data_idx]):  # type: ignore[arg-type]
5209
                raise ValueError(f"Wrong length of predict results for data {data_idx}")
wxchan's avatar
wxchan committed
5210
            self.__is_predicted_cur_iter[data_idx] = True
5211
        result: np.ndarray = self.__inner_predict_buffer[data_idx]  # type: ignore[assignment]
5212
5213
        if self.__num_class > 1:
            num_data = result.size // self.__num_class
5214
            result = result.reshape(num_data, self.__num_class, order="F")
5215
        return result
wxchan's avatar
wxchan committed
5216

5217
    def __get_eval_info(self) -> None:
5218
        """Get inner evaluation count and names."""
wxchan's avatar
wxchan committed
5219
5220
        if self.__need_reload_eval_info:
            self.__need_reload_eval_info = False
Guolin Ke's avatar
Guolin Ke committed
5221
            out_num_eval = ctypes.c_int(0)
5222
            # Get num of inner evals
5223
5224
5225
5226
5227
5228
            _safe_call(
                _LIB.LGBM_BoosterGetEvalCounts(
                    self._handle,
                    ctypes.byref(out_num_eval),
                )
            )
wxchan's avatar
wxchan committed
5229
5230
            self.__num_inner_eval = out_num_eval.value
            if self.__num_inner_eval > 0:
5231
                # Get name of eval metrics
Guolin Ke's avatar
Guolin Ke committed
5232
                tmp_out_len = ctypes.c_int(0)
5233
5234
5235
                reserved_string_buffer_size = 255
                required_string_buffer_size = ctypes.c_size_t(0)
                string_buffers = [
5236
                    ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(self.__num_inner_eval)
5237
                ]
5238
                ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
                _safe_call(
                    _LIB.LGBM_BoosterGetEvalNames(
                        self._handle,
                        ctypes.c_int(self.__num_inner_eval),
                        ctypes.byref(tmp_out_len),
                        ctypes.c_size_t(reserved_string_buffer_size),
                        ctypes.byref(required_string_buffer_size),
                        ptr_string_buffers,
                    )
                )
wxchan's avatar
wxchan committed
5249
                if self.__num_inner_eval != tmp_out_len.value:
5250
                    raise ValueError("Length of eval names doesn't equal with num_evals")
5251
5252
5253
5254
5255
5256
                actual_string_buffer_size = required_string_buffer_size.value
                # if buffer length is not long enough, reallocate buffers
                if reserved_string_buffer_size < actual_string_buffer_size:
                    string_buffers = [
                        ctypes.create_string_buffer(actual_string_buffer_size) for _ in range(self.__num_inner_eval)
                    ]
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
                    ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(
                        *map(ctypes.addressof, string_buffers)
                    )  # type: ignore[misc]
                    _safe_call(
                        _LIB.LGBM_BoosterGetEvalNames(
                            self._handle,
                            ctypes.c_int(self.__num_inner_eval),
                            ctypes.byref(tmp_out_len),
                            ctypes.c_size_t(actual_string_buffer_size),
                            ctypes.byref(required_string_buffer_size),
                            ptr_string_buffers,
                        )
                    )
                self.__name_inner_eval = [string_buffers[i].value.decode("utf-8") for i in range(self.__num_inner_eval)]
5271
                self.__higher_better_inner_eval = [
5272
                    name.startswith(("auc", "ndcg@", "map@", "average_precision")) for name in self.__name_inner_eval
5273
                ]