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

4
5
6
7
8
9
10
# This import causes lib_lightgbm.{dll,dylib,so} to be loaded.
# It's intentionally done here, as early as possible, to avoid issues like
# "libgomp.so.1: cannot allocate memory in static TLS block" on aarch64 Linux.
#
# For details, see the "cannot allocate memory in static TLS block" entry in docs/FAQ.rst.
from .libpath import _LIB  # isort: skip

11
import abc
wxchan's avatar
wxchan committed
12
import ctypes
13
import inspect
14
import json
wxchan's avatar
wxchan committed
15
import warnings
16
from collections import OrderedDict
17
from copy import deepcopy
18
from enum import Enum
19
from functools import wraps
20
from os import SEEK_END, environ
21
22
from os.path import getsize
from pathlib import Path
23
from tempfile import NamedTemporaryFile
24
from typing import TYPE_CHECKING, Any, Callable, Dict, Iterable, Iterator, List, Optional, Set, Tuple, Union
wxchan's avatar
wxchan committed
25
26
27
28

import numpy as np
import scipy.sparse

29
from .compat import (
30
    CFFI_INSTALLED,
31
32
33
    PANDAS_INSTALLED,
    PYARROW_INSTALLED,
    arrow_cffi,
34
    arrow_is_boolean,
35
36
37
38
39
40
41
42
43
44
45
46
    arrow_is_floating,
    arrow_is_integer,
    concat,
    pa_Array,
    pa_chunked_array,
    pa_ChunkedArray,
    pa_compute,
    pa_Table,
    pd_CategoricalDtype,
    pd_DataFrame,
    pd_Series,
)
wxchan's avatar
wxchan committed
47

48
49
50
if TYPE_CHECKING:
    from typing import Literal

51
52
53
54
55
56
57
    # typing.TypeGuard was only introduced in Python 3.10
    try:
        from typing import TypeGuard
    except ImportError:
        from typing_extensions import TypeGuard


58
__all__ = [
59
60
61
62
63
64
    "Booster",
    "Dataset",
    "LGBMDeprecationWarning",
    "LightGBMError",
    "register_logger",
    "Sequence",
65
66
]

67
_BoosterHandle = ctypes.c_void_p
68
_DatasetHandle = ctypes.c_void_p
69
70
_ctypes_int_ptr = Union[
    "ctypes._Pointer[ctypes.c_int32]",
71
    "ctypes._Pointer[ctypes.c_int64]",
72
]
73
74
_ctypes_int_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_int32]]",
75
    "ctypes.Array[ctypes._Pointer[ctypes.c_int64]]",
76
]
77
78
_ctypes_float_ptr = Union[
    "ctypes._Pointer[ctypes.c_float]",
79
    "ctypes._Pointer[ctypes.c_double]",
80
81
82
]
_ctypes_float_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_float]]",
83
    "ctypes.Array[ctypes._Pointer[ctypes.c_double]]",
84
]
85
_LGBM_EvalFunctionResultType = Tuple[str, float, bool]
86
_LGBM_BoosterBestScoreType = Dict[str, Dict[str, float]]
87
_LGBM_BoosterEvalMethodResultType = Tuple[str, str, float, bool]
88
_LGBM_BoosterEvalMethodResultWithStandardDeviationType = Tuple[str, str, float, bool, float]
89
90
_LGBM_CategoricalFeatureConfiguration = Union[List[str], List[int], "Literal['auto']"]
_LGBM_FeatureNameConfiguration = Union[List[str], "Literal['auto']"]
91
92
93
94
_LGBM_GroupType = Union[
    List[float],
    List[int],
    np.ndarray,
95
96
97
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
98
]
99
100
_LGBM_PositionType = Union[
    np.ndarray,
101
    pd_Series,
102
]
103
104
105
106
107
108
_LGBM_InitScoreType = Union[
    List[float],
    List[List[float]],
    np.ndarray,
    pd_Series,
    pd_DataFrame,
109
110
111
    pa_Table,
    pa_Array,
    pa_ChunkedArray,
112
]
113
114
115
116
117
118
119
120
_LGBM_TrainDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    scipy.sparse.spmatrix,
    "Sequence",
    List["Sequence"],
121
    List[np.ndarray],
122
    pa_Table,
123
]
124
_LGBM_LabelType = Union[
125
126
    List[float],
    List[int],
127
128
    np.ndarray,
    pd_Series,
129
130
131
    pd_DataFrame,
    pa_Array,
    pa_ChunkedArray,
132
]
133
134
135
136
137
_LGBM_PredictDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
138
139
    scipy.sparse.spmatrix,
    pa_Table,
140
]
141
142
143
144
_LGBM_WeightType = Union[
    List[float],
    List[int],
    np.ndarray,
145
146
147
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
148
]
149
150
151
152
153
154
155
156
157
158
159
160
161
_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,
]

162
163
ZERO_THRESHOLD = 1e-35

164
165
_MULTICLASS_OBJECTIVES = {"multiclass", "multiclassova", "multiclass_ova", "ova", "ovr", "softmax"}

166

167
168
169
170
171
172
class LightGBMError(Exception):
    """Error thrown by LightGBM."""

    pass


173
174
175
176
def _is_zero(x: float) -> bool:
    return -ZERO_THRESHOLD <= x <= ZERO_THRESHOLD


177
def _get_sample_count(total_nrow: int, params: str) -> int:
178
    sample_cnt = ctypes.c_int(0)
179
180
181
182
183
184
185
    _safe_call(
        _LIB.LGBM_GetSampleCount(
            ctypes.c_int32(total_nrow),
            _c_str(params),
            ctypes.byref(sample_cnt),
        )
    )
186
187
    return sample_cnt.value

wxchan's avatar
wxchan committed
188

189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
def _np2d_to_np1d(mat: np.ndarray) -> Tuple[np.ndarray, int]:
    if mat.dtype in (np.float32, np.float64):
        dtype = mat.dtype
    else:
        dtype = np.float32
    if mat.flags["F_CONTIGUOUS"]:
        order = "F"
        layout = _C_API_IS_COL_MAJOR
    else:
        order = "C"
        layout = _C_API_IS_ROW_MAJOR
    # ensure dtype and order, copies if either do not match
    data = np.asarray(mat, dtype=dtype, order=order)
    # flatten array without copying
    return data.ravel(order=order), layout


206
class _MissingType(Enum):
207
208
209
    NONE = "None"
    NAN = "NaN"
    ZERO = "Zero"
210
211


212
class _DummyLogger:
213
    def info(self, msg: str) -> None:
214
        print(msg)  # noqa: T201
215

216
    def warning(self, msg: str) -> None:
217
218
219
        warnings.warn(msg, stacklevel=3)


220
221
222
_LOGGER: Any = _DummyLogger()
_INFO_METHOD_NAME = "info"
_WARNING_METHOD_NAME = "warning"
223
224


225
226
227
228
def _has_method(logger: Any, method_name: str) -> bool:
    return callable(getattr(logger, method_name, None))


229
def register_logger(
230
231
232
    logger: Any,
    info_method_name: str = "info",
    warning_method_name: str = "warning",
233
) -> None:
234
235
236
237
    """Register custom logger.

    Parameters
    ----------
238
    logger : Any
239
        Custom logger.
240
241
242
243
    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.
244
    """
245
    if not _has_method(logger, info_method_name) or not _has_method(logger, warning_method_name):
246
        raise TypeError(f"Logger must provide '{info_method_name}' and '{warning_method_name}' method")
247
248

    global _LOGGER, _INFO_METHOD_NAME, _WARNING_METHOD_NAME
249
    _LOGGER = logger
250
251
    _INFO_METHOD_NAME = info_method_name
    _WARNING_METHOD_NAME = warning_method_name
252
253


254
def _normalize_native_string(func: Callable[[str], None]) -> Callable[[str], None]:
255
    """Join log messages from native library which come by chunks."""
256
    msg_normalized: List[str] = []
257
258

    @wraps(func)
259
    def wrapper(msg: str) -> None:
260
        nonlocal msg_normalized
261
262
        if msg.strip() == "":
            msg = "".join(msg_normalized)
263
264
265
266
267
268
269
270
            msg_normalized = []
            return func(msg)
        else:
            msg_normalized.append(msg)

    return wrapper


271
def _log_info(msg: str) -> None:
272
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
273
274


275
def _log_warning(msg: str) -> None:
276
    getattr(_LOGGER, _WARNING_METHOD_NAME)(msg)
277
278
279


@_normalize_native_string
280
def _log_native(msg: str) -> None:
281
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
282
283


284
def _log_callback(msg: bytes) -> None:
285
    """Redirect logs from native library into Python."""
286
    _log_native(str(msg.decode("utf-8")))
287
288


289
290
291
# connect the Python logger to logging in lib_lightgbm
if not environ.get("LIGHTGBM_BUILD_DOC", False):
    _LIB.LGBM_GetLastError.restype = ctypes.c_char_p
292
    callback = ctypes.CFUNCTYPE(None, ctypes.c_char_p)
293
294
295
    _LIB.callback = callback(_log_callback)  # type: ignore[attr-defined]
    if _LIB.LGBM_RegisterLogCallback(_LIB.callback) != 0:
        raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
wxchan's avatar
wxchan committed
296

wxchan's avatar
wxchan committed
297

298
_NUMERIC_TYPES = (int, float, bool)
299
300


301
def _safe_call(ret: int) -> None:
302
303
    """Check the return value from C API call.

wxchan's avatar
wxchan committed
304
305
306
    Parameters
    ----------
    ret : int
307
        The return value from C API calls.
wxchan's avatar
wxchan committed
308
309
    """
    if ret != 0:
310
        raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
wxchan's avatar
wxchan committed
311

wxchan's avatar
wxchan committed
312

313
def _is_numeric(obj: Any) -> bool:
314
    """Check whether object is a number or not, include numpy number, etc."""
wxchan's avatar
wxchan committed
315
316
317
    try:
        float(obj)
        return True
wxchan's avatar
wxchan committed
318
319
320
    except (TypeError, ValueError):
        # TypeError: obj is not a string or a number
        # ValueError: invalid literal
wxchan's avatar
wxchan committed
321
322
        return False

wxchan's avatar
wxchan committed
323

324
def _is_numpy_1d_array(data: Any) -> bool:
325
    """Check whether data is a numpy 1-D array."""
326
    return isinstance(data, np.ndarray) and len(data.shape) == 1
wxchan's avatar
wxchan committed
327

wxchan's avatar
wxchan committed
328

329
def _is_numpy_column_array(data: Any) -> bool:
330
331
332
333
334
335
336
    """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


337
def _cast_numpy_array_to_dtype(array: np.ndarray, dtype: "np.typing.DTypeLike") -> np.ndarray:
338
    """Cast numpy array to given dtype."""
339
340
341
342
343
    if array.dtype == dtype:
        return array
    return array.astype(dtype=dtype, copy=False)


344
def _is_1d_list(data: Any) -> bool:
345
    """Check whether data is a 1-D list."""
346
    return isinstance(data, list) and (not data or _is_numeric(data[0]))
wxchan's avatar
wxchan committed
347

wxchan's avatar
wxchan committed
348

349
def _is_list_of_numpy_arrays(data: Any) -> "TypeGuard[List[np.ndarray]]":
350
    return isinstance(data, list) and all(isinstance(x, np.ndarray) for x in data)
351
352
353


def _is_list_of_sequences(data: Any) -> "TypeGuard[List[Sequence]]":
354
    return isinstance(data, list) and all(isinstance(x, Sequence) for x in data)
355
356


357
358
def _is_1d_collection(data: Any) -> bool:
    """Check whether data is a 1-D collection."""
359
    return _is_numpy_1d_array(data) or _is_numpy_column_array(data) or _is_1d_list(data) or isinstance(data, pd_Series)
360
361


362
363
def _list_to_1d_numpy(
    data: Any,
364
    dtype: "np.typing.DTypeLike",
365
    name: str,
366
) -> np.ndarray:
367
    """Convert data to numpy 1-D array."""
368
    if _is_numpy_1d_array(data):
369
        return _cast_numpy_array_to_dtype(data, dtype)
370
    elif _is_numpy_column_array(data):
371
        _log_warning("Converting column-vector to 1d array")
372
        array = data.ravel()
373
        return _cast_numpy_array_to_dtype(array, dtype)
374
    elif _is_1d_list(data):
375
        return np.asarray(data, dtype=dtype)
376
    elif isinstance(data, pd_Series):
377
        _check_for_bad_pandas_dtypes(data.to_frame().dtypes)
378
        return np.asarray(data, dtype=dtype)  # SparseArray should be supported as well
wxchan's avatar
wxchan committed
379
    else:
380
        raise TypeError(
381
            f"Wrong type({type(data).__name__}) for {name}.\nIt should be list, numpy 1-D array or pandas Series"
382
        )
wxchan's avatar
wxchan committed
383

wxchan's avatar
wxchan committed
384

385
386
387
388
389
390
391
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."""
392
    return isinstance(data, list) and len(data) > 0 and _is_1d_list(data[0])
393
394
395
396


def _is_2d_collection(data: Any) -> bool:
    """Check whether data is a 2-D collection."""
397
    return _is_numpy_2d_array(data) or _is_2d_list(data) or isinstance(data, pd_DataFrame)
398
399


400
def _is_pyarrow_array(data: Any) -> "TypeGuard[Union[pa_Array, pa_ChunkedArray]]":
401
402
403
404
    """Check whether data is a PyArrow array."""
    return isinstance(data, (pa_Array, pa_ChunkedArray))


405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
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
436
437
438
439
440
    if isinstance(data, pa_Array):
        export_objects = [data]
    elif isinstance(data, pa_ChunkedArray):
        export_objects = data.chunks
    elif isinstance(data, pa_Table):
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
        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)


461
462
def _data_to_2d_numpy(
    data: Any,
463
    dtype: "np.typing.DTypeLike",
464
    name: str,
465
) -> np.ndarray:
466
467
    """Convert data to numpy 2-D array."""
    if _is_numpy_2d_array(data):
468
        return _cast_numpy_array_to_dtype(data, dtype)
469
470
471
    if _is_2d_list(data):
        return np.array(data, dtype=dtype)
    if isinstance(data, pd_DataFrame):
472
        _check_for_bad_pandas_dtypes(data.dtypes)
473
        return _cast_numpy_array_to_dtype(data.values, dtype)
474
475
476
477
    raise TypeError(
        f"Wrong type({type(data).__name__}) for {name}.\n"
        "It should be list of lists, numpy 2-D array or pandas DataFrame"
    )
478
479


480
def _cfloat32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
481
    """Convert a ctypes float pointer array to a numpy array."""
wxchan's avatar
wxchan committed
482
    if isinstance(cptr, ctypes.POINTER(ctypes.c_float)):
483
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
484
    else:
485
        raise RuntimeError("Expected float pointer")
wxchan's avatar
wxchan committed
486

Guolin Ke's avatar
Guolin Ke committed
487

488
def _cfloat64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
489
    """Convert a ctypes double pointer array to a numpy array."""
Guolin Ke's avatar
Guolin Ke committed
490
    if isinstance(cptr, ctypes.POINTER(ctypes.c_double)):
491
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
Guolin Ke's avatar
Guolin Ke committed
492
    else:
493
        raise RuntimeError("Expected double pointer")
Guolin Ke's avatar
Guolin Ke committed
494

wxchan's avatar
wxchan committed
495

496
def _cint32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
497
    """Convert a ctypes int pointer array to a numpy array."""
wxchan's avatar
wxchan committed
498
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int32)):
499
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
500
    else:
501
        raise RuntimeError("Expected int32 pointer")
502
503


504
def _cint64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
505
506
    """Convert a ctypes int pointer array to a numpy array."""
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int64)):
507
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
508
    else:
509
        raise RuntimeError("Expected int64 pointer")
wxchan's avatar
wxchan committed
510

wxchan's avatar
wxchan committed
511

512
def _c_str(string: str) -> ctypes.c_char_p:
513
    """Convert a Python string to C string."""
514
    return ctypes.c_char_p(string.encode("utf-8"))
wxchan's avatar
wxchan committed
515

wxchan's avatar
wxchan committed
516

517
def _c_array(ctype: type, values: List[Any]) -> ctypes.Array:
518
    """Convert a Python array to C array."""
519
    return (ctype * len(values))(*values)  # type: ignore[operator]
wxchan's avatar
wxchan committed
520

wxchan's avatar
wxchan committed
521

522
def _json_default_with_numpy(obj: Any) -> Any:
523
524
525
526
527
528
529
530
531
    """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


532
533
534
535
536
537
538
539
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)


540
def _param_dict_to_str(data: Optional[Dict[str, Any]]) -> str:
541
    """Convert Python dictionary to string, which is passed to C API."""
542
    if data is None or not data:
wxchan's avatar
wxchan committed
543
544
545
        return ""
    pairs = []
    for key, val in data.items():
546
        if isinstance(val, (list, tuple, set)) or _is_numpy_1d_array(val):
547
            pairs.append(f"{key}={','.join(map(_to_string, val))}")
548
        elif isinstance(val, (str, Path, _NUMERIC_TYPES)) or _is_numeric(val):
549
            pairs.append(f"{key}={val}")
550
        elif val is not None:
551
552
            raise TypeError(f"Unknown type of parameter:{key}, got:{type(val).__name__}")
    return " ".join(pairs)
553

wxchan's avatar
wxchan committed
554

555
class _TempFile:
556
557
    """Proxy class to workaround errors on Windows."""

558
    def __enter__(self) -> "_TempFile":
559
560
        with NamedTemporaryFile(prefix="lightgbm_tmp_", delete=True) as f:
            self.name = f.name
561
            self.path = Path(self.name)
562
        return self
wxchan's avatar
wxchan committed
563

564
    def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None:
565
566
        if self.path.is_file():
            self.path.unlink()
567

wxchan's avatar
wxchan committed
568

569
# DeprecationWarning is not shown by default, so let's create our own with higher level
570
571
# ref: https://peps.python.org/pep-0565/#additional-use-case-for-futurewarning
class LGBMDeprecationWarning(FutureWarning):
572
573
574
575
576
577
    """Custom deprecation warning."""

    pass


class _ConfigAliases:
578
579
580
581
    # lazy evaluation to allow import without dynamic library, e.g., for docs generation
    aliases = None

    @staticmethod
582
    def _get_all_param_aliases() -> Dict[str, List[str]]:
583
584
585
        buffer_len = 1 << 20
        tmp_out_len = ctypes.c_int64(0)
        string_buffer = ctypes.create_string_buffer(buffer_len)
586
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
587
588
589
590
591
592
593
        _safe_call(
            _LIB.LGBM_DumpParamAliases(
                ctypes.c_int64(buffer_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer,
            )
        )
594
595
596
597
        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)
598
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
599
600
601
602
603
604
605
            _safe_call(
                _LIB.LGBM_DumpParamAliases(
                    ctypes.c_int64(actual_len),
                    ctypes.byref(tmp_out_len),
                    ptr_string_buffer,
                )
            )
606
        return json.loads(
607
            string_buffer.value.decode("utf-8"), object_hook=lambda obj: {k: [k] + v for k, v in obj.items()}
608
        )
609
610

    @classmethod
611
    def get(cls, *args: str) -> Set[str]:
612
613
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
614
615
        ret = set()
        for i in args:
616
            ret.update(cls.get_sorted(i))
617
618
        return ret

619
620
621
622
623
624
    @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])

625
    @classmethod
626
    def get_by_alias(cls, *args: str) -> Set[str]:
627
628
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
629
630
631
632
        ret = set(args)
        for arg in args:
            for aliases in cls.aliases.values():
                if arg in aliases:
633
                    ret.update(aliases)
634
635
636
                    break
        return ret

637

638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
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)

659
660
    aliases = _ConfigAliases.get_sorted(main_param_name)
    aliases = [a for a in aliases if a != main_param_name]
661
662

    # if main_param_name was provided, keep that value and remove all aliases
663
    if main_param_name in params.keys():
664
665
666
        for param in aliases:
            params.pop(param, None)
        return params
667

668
669
670
671
672
    # 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
673

674
675
676
677
678
679
680
    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
681
682
683
684

    return params


685
_MAX_INT32 = (1 << 31) - 1
686

687
"""Macro definition of data type in C API of LightGBM"""
688
689
690
691
_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
692

693
694
"""Macro definition of data order in matrix"""
_C_API_IS_COL_MAJOR = 0
695
_C_API_IS_ROW_MAJOR = 1
wxchan's avatar
wxchan committed
696

697
"""Macro definition of prediction type in C API of LightGBM"""
698
699
700
701
_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
702

703
"""Macro definition of sparse matrix type"""
704
705
_C_API_MATRIX_TYPE_CSR = 0
_C_API_MATRIX_TYPE_CSC = 1
706

707
"""Macro definition of feature importance type"""
708
709
_C_API_FEATURE_IMPORTANCE_SPLIT = 0
_C_API_FEATURE_IMPORTANCE_GAIN = 1
710

711
"""Data type of data field"""
712
713
714
715
_FIELD_TYPE_MAPPER = {
    "label": _C_API_DTYPE_FLOAT32,
    "weight": _C_API_DTYPE_FLOAT32,
    "init_score": _C_API_DTYPE_FLOAT64,
716
    "group": _C_API_DTYPE_INT32,
717
    "position": _C_API_DTYPE_INT32,
718
}
wxchan's avatar
wxchan committed
719

720
"""String name to int feature importance type mapper"""
721
722
_FEATURE_IMPORTANCE_TYPE_MAPPER = {
    "split": _C_API_FEATURE_IMPORTANCE_SPLIT,
723
    "gain": _C_API_FEATURE_IMPORTANCE_GAIN,
724
}
725

wxchan's avatar
wxchan committed
726

727
def _convert_from_sliced_object(data: np.ndarray) -> np.ndarray:
728
    """Fix the memory of multi-dimensional sliced object."""
729
    if isinstance(data, np.ndarray) and isinstance(data.base, np.ndarray):
730
        if not data.flags.c_contiguous:
731
732
733
734
            _log_warning(
                "Usage of np.ndarray subset (sliced data) is not recommended "
                "due to it will double the peak memory cost in LightGBM."
            )
735
736
737
738
            return np.copy(data)
    return data


739
def _c_float_array(data: np.ndarray) -> Tuple[_ctypes_float_ptr, int, np.ndarray]:
740
    """Get pointer of float numpy array / list."""
741
    if _is_1d_list(data):
742
        data = np.asarray(data)
743
    if _is_numpy_1d_array(data):
744
        data = _convert_from_sliced_object(data)
745
        assert data.flags.c_contiguous
746
        ptr_data: _ctypes_float_ptr
wxchan's avatar
wxchan committed
747
748
        if data.dtype == np.float32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
749
            type_data = _C_API_DTYPE_FLOAT32
wxchan's avatar
wxchan committed
750
751
        elif data.dtype == np.float64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_double))
752
            type_data = _C_API_DTYPE_FLOAT64
wxchan's avatar
wxchan committed
753
        else:
754
            raise TypeError(f"Expected np.float32 or np.float64, met type({data.dtype})")
wxchan's avatar
wxchan committed
755
    else:
756
        raise TypeError(f"Unknown type({type(data).__name__})")
757
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
758

wxchan's avatar
wxchan committed
759

760
def _c_int_array(data: np.ndarray) -> Tuple[_ctypes_int_ptr, int, np.ndarray]:
761
    """Get pointer of int numpy array / list."""
762
    if _is_1d_list(data):
763
        data = np.asarray(data)
764
    if _is_numpy_1d_array(data):
765
        data = _convert_from_sliced_object(data)
766
        assert data.flags.c_contiguous
767
        ptr_data: _ctypes_int_ptr
wxchan's avatar
wxchan committed
768
769
        if data.dtype == np.int32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int32))
770
            type_data = _C_API_DTYPE_INT32
wxchan's avatar
wxchan committed
771
772
        elif data.dtype == np.int64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int64))
773
            type_data = _C_API_DTYPE_INT64
wxchan's avatar
wxchan committed
774
        else:
775
            raise TypeError(f"Expected np.int32 or np.int64, met type({data.dtype})")
wxchan's avatar
wxchan committed
776
    else:
777
        raise TypeError(f"Unknown type({type(data).__name__})")
778
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
779

wxchan's avatar
wxchan committed
780

781
def _is_allowed_numpy_dtype(dtype: type) -> bool:
782
783
    float128 = getattr(np, "float128", type(None))
    return issubclass(dtype, (np.integer, np.floating, np.bool_)) and not issubclass(dtype, (np.timedelta64, float128))
784
785


786
def _check_for_bad_pandas_dtypes(pandas_dtypes_series: pd_Series) -> None:
787
    bad_pandas_dtypes = [
788
        f"{column_name}: {pandas_dtype}"
789
        for column_name, pandas_dtype in pandas_dtypes_series.items()
790
        if not _is_allowed_numpy_dtype(pandas_dtype.type)
791
792
    ]
    if bad_pandas_dtypes:
793
        raise ValueError(
794
            f"pandas dtypes must be int, float or bool.\nFields with bad pandas dtypes: {', '.join(bad_pandas_dtypes)}"
795
        )
796
797


798
799
def _pandas_to_numpy(
    data: pd_DataFrame,
800
    target_dtype: "np.typing.DTypeLike",
801
802
803
804
805
806
807
808
809
810
811
812
813
814
) -> 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)


815
def _data_from_pandas(
816
817
818
    data: pd_DataFrame,
    feature_name: _LGBM_FeatureNameConfiguration,
    categorical_feature: _LGBM_CategoricalFeatureConfiguration,
819
    pandas_categorical: Optional[List[List]],
820
) -> Tuple[np.ndarray, List[str], Union[List[str], List[int]], List[List]]:
821
    if len(data.shape) != 2 or data.shape[0] < 1:
822
        raise ValueError("Input data must be 2 dimensional and non empty.")
823

824
825
826
827
    # take shallow copy in case we modify categorical columns
    # whole column modifications don't change the original df
    data = data.copy(deep=False)

828
    # determine feature names
829
    if feature_name == "auto":
830
831
832
833
        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)]
834
    cat_cols_not_ordered: List[str] = [col for col in cat_cols if not data[col].cat.ordered]
835
836
    if pandas_categorical is None:  # train dataset
        pandas_categorical = [list(data[col].cat.categories) for col in cat_cols]
837
    else:
838
        if len(cat_cols) != len(pandas_categorical):
839
            raise ValueError("train and valid dataset categorical_feature do not match.")
840
841
842
843
844
        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})
845
846

    # use cat cols from DataFrame
847
    if categorical_feature == "auto":
848
849
850
        categorical_feature = cat_cols_not_ordered

    df_dtypes = [dtype.type for dtype in data.dtypes]
851
852
    # so that the target dtype considers floats
    df_dtypes.append(np.float32)
853
    target_dtype = np.result_type(*df_dtypes)
854
855
856
857
858

    return (
        _pandas_to_numpy(data, target_dtype=target_dtype),
        feature_name,
        categorical_feature,
859
        pandas_categorical,
860
    )
861
862


863
864
def _dump_pandas_categorical(
    pandas_categorical: Optional[List[List]],
865
    file_name: Optional[Union[str, Path]] = None,
866
) -> str:
867
    categorical_json = json.dumps(pandas_categorical, default=_json_default_with_numpy)
868
    pandas_str = f"\npandas_categorical:{categorical_json}\n"
869
    if file_name is not None:
870
        with open(file_name, "a") as f:
871
872
873
874
            f.write(pandas_str)
    return pandas_str


875
876
def _load_pandas_categorical(
    file_name: Optional[Union[str, Path]] = None,
877
    model_str: Optional[str] = None,
878
) -> Optional[List[List]]:
879
    pandas_key = "pandas_categorical:"
880
    offset = -len(pandas_key)
881
    if file_name is not None:
882
        max_offset = -getsize(file_name)
883
        with open(file_name, "rb") as f:
884
            while True:
885
                offset = max(offset, max_offset)
886
                f.seek(offset, SEEK_END)
887
888
889
890
                lines = f.readlines()
                if len(lines) >= 2:
                    break
                offset *= 2
891
        last_line = lines[-1].decode("utf-8").strip()
892
        if not last_line.startswith(pandas_key):
893
            last_line = lines[-2].decode("utf-8").strip()
894
    elif model_str is not None:
895
        idx = model_str.rfind("\n", 0, offset)
896
897
        last_line = model_str[idx:].strip()
    if last_line.startswith(pandas_key):
898
        return json.loads(last_line[len(pandas_key) :])
899
900
    else:
        return None
901
902


903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
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**.

923
924
    .. versionadded:: 3.3.0

925
926
927
928
929
930
931
932
933
    Attributes
    ----------
    batch_size : int
        Default size of a batch.
    """

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

    @abc.abstractmethod
934
    def __getitem__(self, idx: Union[int, slice, List[int]]) -> np.ndarray:
935
936
937
938
939
940
941
        """Return data for given row index.

        A basic implementation should look like this:

        .. code-block:: python

            if isinstance(idx, numbers.Integral):
942
                return self._get_one_line(idx)
943
            elif isinstance(idx, slice):
944
945
                return np.stack([self._get_one_line(i) for i in range(idx.start, idx.stop)])
            elif isinstance(idx, list):
946
                # Only required if using ``Dataset.subset()``.
947
                return np.array([self._get_one_line(i) for i in idx])
948
            else:
949
                raise TypeError(f"Sequence index must be integer, slice or list, got {type(idx).__name__}")
950
951
952

        Parameters
        ----------
953
        idx : int, slice[int], list[int]
954
955
956
957
            Item index.

        Returns
        -------
958
        result : numpy 1-D array or numpy 2-D array
959
            1-D array if idx is int, 2-D array if idx is slice or list.
960
961
962
963
964
965
966
967
968
        """
        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__()")


969
class _InnerPredictor:
970
971
972
973
974
    """_InnerPredictor of LightGBM.

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

Nikita Titov's avatar
Nikita Titov committed
975
976
977
    .. note::

        Can be converted from Booster, but cannot be converted to Booster.
Guolin Ke's avatar
Guolin Ke committed
978
    """
979

980
981
    def __init__(
        self,
982
983
984
        booster_handle: _BoosterHandle,
        pandas_categorical: Optional[List[List]],
        pred_parameter: Dict[str, Any],
985
        manage_handle: bool,
986
    ):
987
        """Initialize the _InnerPredictor.
wxchan's avatar
wxchan committed
988
989
990

        Parameters
        ----------
991
        booster_handle : object
992
            Handle of Booster.
993
994
995
996
        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
997
            Other parameters for the prediction.
998
999
        manage_handle : bool
            If ``True``, free the corresponding Booster on the C++ side when this Python object is deleted.
wxchan's avatar
wxchan committed
1000
        """
1001
1002
1003
1004
1005
1006
1007
1008
        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(
1009
                self._handle,
1010
                ctypes.byref(out_num_class),
1011
1012
1013
            )
        )
        self.num_class = out_num_class.value
wxchan's avatar
wxchan committed
1014

1015
1016
1017
1018
    @classmethod
    def from_booster(
        cls,
        booster: "Booster",
1019
        pred_parameter: Dict[str, Any],
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
    ) -> "_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,
1034
                ctypes.byref(out_cur_iter),
1035
1036
1037
1038
1039
1040
            )
        )
        return cls(
            booster_handle=booster._handle,
            pandas_categorical=booster.pandas_categorical,
            pred_parameter=pred_parameter,
1041
            manage_handle=False,
1042
1043
1044
1045
1046
1047
        )

    @classmethod
    def from_model_file(
        cls,
        model_file: Union[str, Path],
1048
        pred_parameter: Dict[str, Any],
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
    ) -> "_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),
1065
                ctypes.byref(booster_handle),
1066
1067
1068
1069
1070
1071
            )
        )
        return cls(
            booster_handle=booster_handle,
            pandas_categorical=_load_pandas_categorical(file_name=model_file),
            pred_parameter=pred_parameter,
1072
            manage_handle=True,
1073
        )
cbecker's avatar
cbecker committed
1074

1075
    def __del__(self) -> None:
1076
1077
        try:
            if self.__is_manage_handle:
1078
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
1079
1080
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
1081

1082
    def __getstate__(self) -> Dict[str, Any]:
1083
        this = self.__dict__.copy()
1084
1085
        this.pop("handle", None)
        this.pop("_handle", None)
1086
1087
        return this

1088
1089
    def predict(
        self,
1090
        data: _LGBM_PredictDataType,
1091
1092
1093
1094
1095
1096
        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,
1097
        validate_features: bool = False,
1098
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
1099
        """Predict logic.
wxchan's avatar
wxchan committed
1100
1101
1102

        Parameters
        ----------
1103
        data : str, pathlib.Path, numpy array, pandas DataFrame, scipy.sparse or pyarrow Table
1104
            Data source for prediction.
1105
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
1106
1107
        start_iteration : int, optional (default=0)
            Start index of the iteration to predict.
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
        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.
1119
1120
1121
        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
1122

1123
1124
            .. versionadded:: 4.0.0

wxchan's avatar
wxchan committed
1125
1126
        Returns
        -------
1127
        result : numpy array, scipy.sparse or list of scipy.sparse
1128
            Prediction result.
1129
            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
1130
        """
wxchan's avatar
wxchan committed
1131
        if isinstance(data, Dataset):
1132
            raise TypeError("Cannot use Dataset instance for prediction, please use raw data instead")
1133
        if isinstance(data, pd_DataFrame) and validate_features:
1134
1135
            data_names = [str(x) for x in data.columns]
            ptr_names = (ctypes.c_char_p * len(data_names))()
1136
            ptr_names[:] = [x.encode("utf-8") for x in data_names]
1137
1138
            _safe_call(
                _LIB.LGBM_BoosterValidateFeatureNames(
1139
                    self._handle,
1140
1141
1142
1143
                    ptr_names,
                    ctypes.c_int(len(data_names)),
                )
            )
1144
1145
1146
1147
1148
1149

        if isinstance(data, pd_DataFrame):
            data = _data_from_pandas(
                data=data,
                feature_name="auto",
                categorical_feature="auto",
1150
                pandas_categorical=self.pandas_categorical,
1151
1152
            )[0]

1153
        predict_type = _C_API_PREDICT_NORMAL
wxchan's avatar
wxchan committed
1154
        if raw_score:
1155
            predict_type = _C_API_PREDICT_RAW_SCORE
wxchan's avatar
wxchan committed
1156
        if pred_leaf:
1157
            predict_type = _C_API_PREDICT_LEAF_INDEX
1158
        if pred_contrib:
1159
            predict_type = _C_API_PREDICT_CONTRIB
cbecker's avatar
cbecker committed
1160

1161
        if isinstance(data, (str, Path)):
1162
            with _TempFile() as f:
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
                _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),
                    )
                )
1175
1176
                preds = np.loadtxt(f.name, dtype=np.float64)
                nrow = preds.shape[0]
wxchan's avatar
wxchan committed
1177
        elif isinstance(data, scipy.sparse.csr_matrix):
1178
1179
1180
1181
            preds, nrow = self.__pred_for_csr(
                csr=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1182
                predict_type=predict_type,
1183
            )
Guolin Ke's avatar
Guolin Ke committed
1184
        elif isinstance(data, scipy.sparse.csc_matrix):
1185
1186
1187
1188
            preds, nrow = self.__pred_for_csc(
                csc=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1189
                predict_type=predict_type,
1190
            )
wxchan's avatar
wxchan committed
1191
        elif isinstance(data, np.ndarray):
1192
1193
1194
1195
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1196
                predict_type=predict_type,
1197
            )
1198
1199
1200
1201
1202
        elif _is_pyarrow_table(data):
            preds, nrow = self.__pred_for_pyarrow_table(
                table=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1203
                predict_type=predict_type,
1204
            )
1205
1206
1207
        elif isinstance(data, list):
            try:
                data = np.array(data)
1208
            except BaseException as err:
1209
                raise ValueError("Cannot convert data list to numpy array.") from err
1210
1211
1212
1213
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1214
                predict_type=predict_type,
1215
            )
wxchan's avatar
wxchan committed
1216
1217
        else:
            try:
1218
                _log_warning("Converting data to scipy sparse matrix.")
wxchan's avatar
wxchan committed
1219
                csr = scipy.sparse.csr_matrix(data)
1220
            except BaseException as err:
1221
                raise TypeError(f"Cannot predict data for type {type(data).__name__}") from err
1222
1223
1224
1225
            preds, nrow = self.__pred_for_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1226
                predict_type=predict_type,
1227
            )
wxchan's avatar
wxchan committed
1228
1229
        if pred_leaf:
            preds = preds.astype(np.int32)
1230
        is_sparse = isinstance(preds, (list, scipy.sparse.spmatrix))
1231
        if not is_sparse and (preds.size != nrow or pred_leaf or pred_contrib):
wxchan's avatar
wxchan committed
1232
            if preds.size % nrow == 0:
1233
                preds = preds.reshape(nrow, -1)
wxchan's avatar
wxchan committed
1234
            else:
1235
                raise ValueError(f"Length of predict result ({preds.size}) cannot be divide nrow ({nrow})")
wxchan's avatar
wxchan committed
1236
1237
        return preds

1238
1239
1240
1241
1242
    def __get_num_preds(
        self,
        start_iteration: int,
        num_iteration: int,
        nrow: int,
1243
        predict_type: int,
1244
    ) -> int:
1245
        """Get size of prediction result."""
1246
        if nrow > _MAX_INT32:
1247
1248
1249
1250
1251
1252
            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
1253
        n_preds = ctypes.c_int64(0)
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
        _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
1264
        return n_preds.value
wxchan's avatar
wxchan committed
1265

1266
1267
1268
1269
1270
1271
    def __inner_predict_np2d(
        self,
        mat: np.ndarray,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
1272
        preds: Optional[np.ndarray],
1273
    ) -> Tuple[np.ndarray, int]:
1274
        data, layout = _np2d_to_np1d(mat)
1275
        ptr_data, type_ptr_data, _ = _c_float_array(data)
1276
1277
1278
1279
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=mat.shape[0],
1280
            predict_type=predict_type,
1281
        )
1282
1283
1284
1285
1286
        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)
1287
1288
1289
1290
1291
1292
1293
        _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]),
1294
                ctypes.c_int(layout),
1295
1296
1297
1298
1299
1300
1301
1302
                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)),
            )
        )
1303
1304
1305
1306
1307
1308
1309
1310
1311
        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,
1312
        predict_type: int,
1313
    ) -> Tuple[np.ndarray, int]:
1314
        """Predict for a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
1315
        if len(mat.shape) != 2:
1316
            raise ValueError("Input numpy.ndarray or list must be 2 dimensional")
wxchan's avatar
wxchan committed
1317

1318
        nrow = mat.shape[0]
1319
1320
        if nrow > _MAX_INT32:
            sections = np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)
1321
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1322
1323
1324
1325
            n_preds = [
                self.__get_num_preds(start_iteration, num_iteration, i, predict_type)
                for i in np.diff([0] + list(sections) + [nrow])
            ]
1326
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1327
            preds = np.empty(sum(n_preds), dtype=np.float64)
1328
1329
1330
            for chunk, (start_idx_pred, end_idx_pred) in zip(
                np.array_split(mat, sections), zip(n_preds_sections, n_preds_sections[1:])
            ):
1331
                # avoid memory consumption by arrays concatenation operations
1332
1333
1334
1335
1336
                self.__inner_predict_np2d(
                    mat=chunk,
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
1337
                    preds=preds[start_idx_pred:end_idx_pred],
1338
                )
1339
            return preds, nrow
wxchan's avatar
wxchan committed
1340
        else:
1341
1342
1343
1344
1345
            return self.__inner_predict_np2d(
                mat=mat,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
1346
                preds=None,
1347
            )
wxchan's avatar
wxchan committed
1348

1349
1350
1351
    def __create_sparse_native(
        self,
        cs: Union[scipy.sparse.csc_matrix, scipy.sparse.csr_matrix],
1352
1353
1354
1355
1356
1357
        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,
1358
        is_csr: bool,
1359
    ) -> Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]]:
1360
1361
1362
        # create numpy array from output arrays
        data_indices_len = out_shape[0]
        indptr_len = out_shape[1]
1363
        if indptr_type == _C_API_DTYPE_INT32:
1364
            out_indptr = _cint32_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1365
        elif indptr_type == _C_API_DTYPE_INT64:
1366
            out_indptr = _cint64_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1367
1368
        else:
            raise TypeError("Expected int32 or int64 type for indptr")
1369
        if data_type == _C_API_DTYPE_FLOAT32:
1370
            out_data = _cfloat32_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1371
        elif data_type == _C_API_DTYPE_FLOAT64:
1372
            out_data = _cfloat64_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1373
1374
        else:
            raise TypeError("Expected float32 or float64 type for data")
1375
        out_indices = _cint32_array_to_numpy(cptr=out_ptr_indices, length=data_indices_len)
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
        # 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]
1387
1388
            cs_indices = out_indices[offset + cs_indptr[0] : offset + matrix_indptr_len]
            cs_data = out_data[offset + cs_indptr[0] : offset + matrix_indptr_len]
1389
1390
1391
1392
1393
1394
1395
1396
1397
            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
1398
1399
1400
1401
1402
1403
1404
1405
1406
        _safe_call(
            _LIB.LGBM_BoosterFreePredictSparse(
                out_ptr_indptr,
                out_ptr_indices,
                out_ptr_data,
                ctypes.c_int(indptr_type),
                ctypes.c_int(data_type),
            )
        )
1407
1408
1409
1410
        if len(cs_output_matrices) == 1:
            return cs_output_matrices[0]
        return cs_output_matrices

1411
1412
1413
1414
1415
1416
    def __inner_predict_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
1417
        preds: Optional[np.ndarray],
1418
1419
    ) -> Tuple[np.ndarray, int]:
        nrow = len(csr.indptr) - 1
1420
1421
1422
1423
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
1424
            predict_type=predict_type,
1425
        )
1426
1427
1428
1429
1430
        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
1431

1432
1433
        ptr_indptr, type_ptr_indptr, _ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
1434

1435
1436
1437
        assert csr.shape[1] <= _MAX_INT32
        csr_indices = csr.indices.astype(np.int32, copy=False)

1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
        _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)),
            )
        )
1457
1458
1459
1460
1461
1462
1463
1464
1465
        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,
1466
        predict_type: int,
1467
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1468
1469
1470
1471
        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
1472
        out_ptr_indptr: _ctypes_int_ptr
1473
1474
1475
1476
1477
        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)()
1478
        out_ptr_data: _ctypes_float_ptr
1479
1480
1481
1482
1483
        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)
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
        _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),
            )
        )
1506
1507
1508
1509
1510
1511
1512
1513
        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,
1514
            is_csr=True,
1515
1516
1517
1518
        )
        nrow = len(csr.indptr) - 1
        return matrices, nrow

1519
1520
1521
1522
1523
    def __pred_for_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
1524
        predict_type: int,
1525
    ) -> Tuple[np.ndarray, int]:
1526
        """Predict for a CSR data."""
1527
        if predict_type == _C_API_PREDICT_CONTRIB:
1528
1529
1530
1531
            return self.__inner_predict_csr_sparse(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1532
                predict_type=predict_type,
1533
            )
1534
        nrow = len(csr.indptr) - 1
1535
1536
        if nrow > _MAX_INT32:
            sections = [0] + list(np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)) + [nrow]
1537
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1538
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff(sections)]
1539
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1540
            preds = np.empty(sum(n_preds), dtype=np.float64)
1541
1542
1543
            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:])
            ):
1544
                # avoid memory consumption by arrays concatenation operations
1545
1546
1547
1548
1549
                self.__inner_predict_csr(
                    csr=csr[start_idx:end_idx],
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
1550
                    preds=preds[start_idx_pred:end_idx_pred],
1551
                )
1552
1553
            return preds, nrow
        else:
1554
1555
1556
1557
1558
            return self.__inner_predict_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
1559
                preds=None,
1560
1561
1562
1563
            )

    def __inner_predict_sparse_csc(
        self,
1564
1565
1566
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
1567
        predict_type: int,
1568
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1569
1570
1571
1572
        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
1573
        out_ptr_indptr: _ctypes_int_ptr
1574
1575
1576
1577
1578
        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)()
1579
        out_ptr_data: _ctypes_float_ptr
1580
1581
1582
1583
1584
        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)
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
        _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),
            )
        )
1607
1608
1609
1610
1611
1612
1613
1614
        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,
1615
            is_csr=False,
1616
1617
1618
        )
        nrow = csc.shape[0]
        return matrices, nrow
Guolin Ke's avatar
Guolin Ke committed
1619

1620
1621
1622
1623
1624
    def __pred_for_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
1625
        predict_type: int,
1626
    ) -> Tuple[np.ndarray, int]:
1627
        """Predict for a CSC data."""
Guolin Ke's avatar
Guolin Ke committed
1628
        nrow = csc.shape[0]
1629
        if nrow > _MAX_INT32:
1630
1631
1632
1633
            return self.__pred_for_csr(
                csr=csc.tocsr(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1634
                predict_type=predict_type,
1635
            )
1636
        if predict_type == _C_API_PREDICT_CONTRIB:
1637
1638
1639
1640
            return self.__inner_predict_sparse_csc(
                csc=csc,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
1641
                predict_type=predict_type,
1642
            )
1643
1644
1645
1646
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
1647
            predict_type=predict_type,
1648
        )
1649
        preds = np.empty(n_preds, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
1650
1651
        out_num_preds = ctypes.c_int64(0)

1652
1653
        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
1654

1655
        assert csc.shape[0] <= _MAX_INT32
1656
        csc_indices = csc.indices.astype(np.int32, copy=False)
1657

1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
        _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
1677
        if n_preds != out_num_preds.value:
1678
            raise ValueError("Wrong length for predict results")
wxchan's avatar
wxchan committed
1679
        return preds, nrow
1680

1681
1682
1683
1684
1685
    def __pred_for_pyarrow_table(
        self,
        table: pa_Table,
        start_iteration: int,
        num_iteration: int,
1686
        predict_type: int,
1687
1688
    ) -> Tuple[np.ndarray, int]:
        """Predict for a PyArrow table."""
1689
1690
        if not (PYARROW_INSTALLED and CFFI_INSTALLED):
            raise LightGBMError("Cannot predict from Arrow without 'pyarrow' and 'cffi' installed.")
1691
1692

        # Check that the input is valid: we only handle numbers (for now)
1693
        if not all(arrow_is_integer(t) or arrow_is_floating(t) or arrow_is_boolean(t) for t in table.schema.types):
1694
1695
1696
1697
1698
1699
1700
            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,
1701
            predict_type=predict_type,
1702
1703
1704
1705
1706
1707
        )
        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)
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
        _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)),
            )
        )
1722
1723
1724
        if n_preds != out_num_preds.value:
            raise ValueError("Wrong length for predict results")
        return preds, table.num_rows
wxchan's avatar
wxchan committed
1725

1726
    def current_iteration(self) -> int:
1727
1728
1729
1730
1731
1732
1733
1734
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
        out_cur_iter = ctypes.c_int(0)
1735
1736
1737
1738
1739
1740
        _safe_call(
            _LIB.LGBM_BoosterGetCurrentIteration(
                self._handle,
                ctypes.byref(out_cur_iter),
            )
        )
1741
1742
        return out_cur_iter.value

wxchan's avatar
wxchan committed
1743

1744
class Dataset:
1745
1746
1747
1748
1749
1750
1751
1752
1753
    """
    Dataset in LightGBM.

    LightGBM does not train on raw data.
    It discretizes continuous features into histogram bins, tries to combine categorical features,
    and automatically handles missing and infinite values.

    This class handles that preprocessing, and holds that alternative representation of the input data.
    """
1754

1755
1756
    def __init__(
        self,
1757
        data: _LGBM_TrainDataType,
1758
        label: Optional[_LGBM_LabelType] = None,
1759
        reference: Optional["Dataset"] = None,
1760
1761
1762
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
1763
1764
        feature_name: _LGBM_FeatureNameConfiguration = "auto",
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = "auto",
1765
        params: Optional[Dict[str, Any]] = None,
1766
1767
        free_raw_data: bool = True,
        position: Optional[_LGBM_PositionType] = None,
1768
    ):
1769
        """Initialize Dataset.
1770

wxchan's avatar
wxchan committed
1771
1772
        Parameters
        ----------
1773
        data : str, pathlib.Path, numpy array, pandas DataFrame, scipy.sparse, Sequence, list of Sequence, list of numpy array or pyarrow Table
wxchan's avatar
wxchan committed
1774
            Data source of Dataset.
1775
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
1776
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1777
1778
1779
            Label of the data.
        reference : Dataset or None, optional (default=None)
            If this is Dataset for validation, training data should be used as reference.
1780
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1781
            Weight for each instance. Weights should be non-negative.
1782
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1783
1784
1785
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
1786
1787
            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.
1788
        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)
1789
            Init score for Dataset.
1790
        feature_name : list of str, or 'auto', optional (default="auto")
1791
            Feature names.
1792
            If 'auto' and data is pandas DataFrame or pyarrow Table, data columns names are used.
1793
        categorical_feature : list of str or int, or 'auto', optional (default="auto")
1794
1795
            Categorical features.
            If list of int, interpreted as indices.
1796
            If list of str, interpreted as feature names (need to specify ``feature_name`` as well).
1797
            If 'auto' and data is pandas DataFrame, pandas unordered categorical columns are used.
1798
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
1799
            Large values could be memory consuming. Consider using consecutive integers starting from zero.
1800
            All negative values in categorical features will be treated as missing values.
1801
            The output cannot be monotonically constrained with respect to a categorical feature.
1802
            Floating point numbers in categorical features will be rounded towards 0.
Nikita Titov's avatar
Nikita Titov committed
1803
        params : dict or None, optional (default=None)
1804
            Other parameters for Dataset.
Nikita Titov's avatar
Nikita Titov committed
1805
        free_raw_data : bool, optional (default=True)
1806
            If True, raw data is freed after constructing inner Dataset.
1807
1808
        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
1809
        """
1810
        self._handle: Optional[_DatasetHandle] = None
wxchan's avatar
wxchan committed
1811
1812
1813
1814
1815
        self.data = data
        self.label = label
        self.reference = reference
        self.weight = weight
        self.group = group
1816
        self.position = position
1817
        self.init_score = init_score
1818
1819
        self.feature_name: _LGBM_FeatureNameConfiguration = feature_name
        self.categorical_feature: _LGBM_CategoricalFeatureConfiguration = categorical_feature
1820
        self.params = deepcopy(params)
wxchan's avatar
wxchan committed
1821
        self.free_raw_data = free_raw_data
1822
        self.used_indices: Optional[List[int]] = None
1823
        self._need_slice = True
1824
        self._predictor: Optional[_InnerPredictor] = None
1825
        self.pandas_categorical: Optional[List[List]] = None
1826
        self._params_back_up: Optional[Dict[str, Any]] = None
1827
        self.version = 0
1828
        self._start_row = 0  # Used when pushing rows one by one.
wxchan's avatar
wxchan committed
1829

1830
    def __del__(self) -> None:
1831
1832
1833
1834
        try:
            self._free_handle()
        except AttributeError:
            pass
1835

1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
    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.
        """
1853
        param_str = _param_dict_to_str(self.get_params())
1854
1855
        sample_cnt = _get_sample_count(total_nrow, param_str)
        indices = np.empty(sample_cnt, dtype=np.int32)
1856
        ptr_data, _, _ = _c_int_array(indices)
1857
1858
        actual_sample_cnt = ctypes.c_int32(0)

1859
1860
1861
1862
1863
1864
1865
1866
        _safe_call(
            _LIB.LGBM_SampleIndices(
                ctypes.c_int32(total_nrow),
                _c_str(param_str),
                ptr_data,
                ctypes.byref(actual_sample_cnt),
            )
        )
1867
1868
        assert sample_cnt == actual_sample_cnt.value
        return indices
1869

1870
1871
1872
    def _init_from_ref_dataset(
        self,
        total_nrow: int,
1873
1874
        ref_dataset: _DatasetHandle,
    ) -> "Dataset":
1875
1876
1877
1878
1879
1880
        """Create dataset from a reference dataset.

        Parameters
        ----------
        total_nrow : int
            Number of rows expected to add to dataset.
1881
1882
        ref_dataset : object
            Handle of reference dataset to extract metadata from.
1883
1884
1885
1886
1887
1888

        Returns
        -------
        self : Dataset
            Constructed Dataset object.
        """
1889
        self._handle = ctypes.c_void_p()
1890
1891
1892
1893
1894
1895
1896
        _safe_call(
            _LIB.LGBM_DatasetCreateByReference(
                ref_dataset,
                ctypes.c_int64(total_nrow),
                ctypes.byref(self._handle),
            )
        )
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
        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
        ----------
1910
        sample_data : list of numpy array
1911
            Sample data for each column.
1912
        sample_indices : list of numpy array
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
            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.
1935
        sample_col_ptr: _ctypes_float_array = (ctypes.POINTER(ctypes.c_double) * ncol)()
1936
1937
        # c type int**
        # each int* points to start of indices for each column
1938
        indices_col_ptr: _ctypes_int_array = (ctypes.POINTER(ctypes.c_int32) * ncol)()
1939
        for i in range(ncol):
1940
1941
            sample_col_ptr[i] = _c_float_array(sample_data[i])[0]
            indices_col_ptr[i] = _c_int_array(sample_indices[i])[0]
1942
1943

        num_per_col = np.array([len(d) for d in sample_indices], dtype=np.int32)
1944
        num_per_col_ptr, _, _ = _c_int_array(num_per_col)
1945

1946
        self._handle = ctypes.c_void_p()
1947
        params_str = _param_dict_to_str(self.get_params())
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
        _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),
            )
        )
1961
1962
        return self

1963
    def _push_rows(self, data: np.ndarray) -> "Dataset":
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
        """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)
1978
        data_ptr, data_type, _ = _c_float_array(data)
1979

1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
        _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),
            )
        )
1990
1991
1992
        self._start_row += nrow
        return self

1993
    def get_params(self) -> Dict[str, Any]:
1994
1995
1996
1997
        """Get the used parameters in the Dataset.

        Returns
        -------
1998
        params : dict
1999
2000
2001
2002
            The used parameters in this Dataset object.
        """
        if self.params is not None:
            # no min_data, nthreads and verbose in this function
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
            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",
            )
2026
            return {k: v for k, v in self.params.items() if k in dataset_params}
2027
2028
        else:
            return {}
2029

2030
    def _free_handle(self) -> "Dataset":
2031
2032
2033
        if self._handle is not None:
            _safe_call(_LIB.LGBM_DatasetFree(self._handle))
            self._handle = None
2034
        self._need_slice = True
Guolin Ke's avatar
Guolin Ke committed
2035
2036
        if self.used_indices is not None:
            self.data = None
Nikita Titov's avatar
Nikita Titov committed
2037
        return self
wxchan's avatar
wxchan committed
2038

2039
2040
2041
    def _set_init_score_by_predictor(
        self,
        predictor: Optional[_InnerPredictor],
2042
        data: _LGBM_TrainDataType,
2043
        used_indices: Optional[Union[List[int], np.ndarray]],
2044
    ) -> "Dataset":
Guolin Ke's avatar
Guolin Ke committed
2045
        data_has_header = False
2046
        if isinstance(data, (str, Path)) and self.params is not None:
Guolin Ke's avatar
Guolin Ke committed
2047
            # check data has header or not
2048
            data_has_header = any(self.params.get(alias, False) for alias in _ConfigAliases.get("header"))
Guolin Ke's avatar
Guolin Ke committed
2049
        num_data = self.num_data()
2050
        if predictor is not None:
2051
2052
2053
            init_score: Union[np.ndarray, scipy.sparse.spmatrix] = predictor.predict(
                data=data,
                raw_score=True,
2054
                data_has_header=data_has_header,
2055
            )
2056
            init_score = init_score.ravel()
2057
            if used_indices is not None:
2058
                assert not self._need_slice
2059
                if isinstance(data, (str, Path)):
2060
                    sub_init_score = np.empty(num_data * predictor.num_class, dtype=np.float64)
2061
                    assert num_data == len(used_indices)
2062
2063
                    for i in range(len(used_indices)):
                        for j in range(predictor.num_class):
2064
2065
2066
                            sub_init_score[i * predictor.num_class + j] = init_score[
                                used_indices[i] * predictor.num_class + j
                            ]
2067
2068
2069
                    init_score = sub_init_score
            if predictor.num_class > 1:
                # need to regroup init_score
2070
                new_init_score = np.empty(init_score.size, dtype=np.float64)
2071
2072
                for i in range(num_data):
                    for j in range(predictor.num_class):
2073
2074
2075
                        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:
2076
            init_score = np.full_like(self.init_score, fill_value=0.0, dtype=np.float64)
2077
2078
        else:
            return self
Guolin Ke's avatar
Guolin Ke committed
2079
        self.set_init_score(init_score)
2080
        return self
Guolin Ke's avatar
Guolin Ke committed
2081

2082
2083
    def _lazy_init(
        self,
2084
        data: Optional[_LGBM_TrainDataType],
2085
2086
2087
2088
2089
2090
2091
2092
        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,
2093
        params: Optional[Dict[str, Any]],
2094
        position: Optional[_LGBM_PositionType],
2095
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2096
        if data is None:
2097
            self._handle = None
Nikita Titov's avatar
Nikita Titov committed
2098
            return self
Guolin Ke's avatar
Guolin Ke committed
2099
2100
2101
        if reference is not None:
            self.pandas_categorical = reference.pandas_categorical
            categorical_feature = reference.categorical_feature
2102
2103
2104
2105
2106
        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,
2107
                pandas_categorical=self.pandas_categorical,
2108
            )
2109
2110
        elif _is_pyarrow_table(data) and feature_name == "auto":
            feature_name = data.column_names
Guolin Ke's avatar
Guolin Ke committed
2111

2112
        # process for args
wxchan's avatar
wxchan committed
2113
        params = {} if params is None else params
2114
        args_names = inspect.signature(self.__class__._lazy_init).parameters.keys()
2115
        for key in params.keys():
2116
            if key in args_names:
2117
2118
2119
2120
                _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."
                )
2121
        # get categorical features
2122
        if isinstance(categorical_feature, list):
2123
2124
            categorical_indices = set()
            feature_dict = {}
2125
            if isinstance(feature_name, list):
2126
2127
                feature_dict = {name: i for i, name in enumerate(feature_name)}
            for name in categorical_feature:
2128
                if isinstance(name, str) and name in feature_dict:
2129
                    categorical_indices.add(feature_dict[name])
2130
                elif isinstance(name, int):
2131
2132
                    categorical_indices.add(name)
                else:
2133
                    raise TypeError(f"Wrong type({type(name).__name__}) or unknown name({name}) in categorical_feature")
2134
            if categorical_indices:
2135
2136
                for cat_alias in _ConfigAliases.get("categorical_feature"):
                    if cat_alias in params:
2137
                        # If the params[cat_alias] is equal to categorical_indices, do not report the warning.
2138
                        if not (isinstance(params[cat_alias], list) and set(params[cat_alias]) == categorical_indices):
2139
                            _log_warning(f"{cat_alias} in param dict is overridden.")
2140
                        params.pop(cat_alias, None)
2141
                params["categorical_column"] = sorted(categorical_indices)
2142

2143
        params_str = _param_dict_to_str(params)
2144
        self.params = params
2145
        # process for reference dataset
wxchan's avatar
wxchan committed
2146
        ref_dataset = None
wxchan's avatar
wxchan committed
2147
        if isinstance(reference, Dataset):
2148
            ref_dataset = reference.construct()._handle
wxchan's avatar
wxchan committed
2149
        elif reference is not None:
2150
            raise TypeError("Reference dataset should be None or dataset instance")
2151
        # start construct data
2152
        if isinstance(data, (str, Path)):
2153
            self._handle = ctypes.c_void_p()
2154
2155
2156
2157
2158
2159
2160
2161
            _safe_call(
                _LIB.LGBM_DatasetCreateFromFile(
                    _c_str(str(data)),
                    _c_str(params_str),
                    ref_dataset,
                    ctypes.byref(self._handle),
                )
            )
wxchan's avatar
wxchan committed
2162
2163
        elif isinstance(data, scipy.sparse.csr_matrix):
            self.__init_from_csr(data, params_str, ref_dataset)
Guolin Ke's avatar
Guolin Ke committed
2164
2165
        elif isinstance(data, scipy.sparse.csc_matrix):
            self.__init_from_csc(data, params_str, ref_dataset)
wxchan's avatar
wxchan committed
2166
2167
        elif isinstance(data, np.ndarray):
            self.__init_from_np2d(data, params_str, ref_dataset)
2168
2169
        elif _is_pyarrow_table(data):
            self.__init_from_pyarrow_table(data, params_str, ref_dataset)
2170
        elif isinstance(data, list) and len(data) > 0:
2171
            if _is_list_of_numpy_arrays(data):
2172
                self.__init_from_list_np2d(data, params_str, ref_dataset)
2173
            elif _is_list_of_sequences(data):
2174
2175
                self.__init_from_seqs(data, ref_dataset)
            else:
2176
                raise TypeError("Data list can only be of ndarray or Sequence")
2177
2178
        elif isinstance(data, Sequence):
            self.__init_from_seqs([data], ref_dataset)
wxchan's avatar
wxchan committed
2179
2180
2181
2182
        else:
            try:
                csr = scipy.sparse.csr_matrix(data)
                self.__init_from_csr(csr, params_str, ref_dataset)
2183
            except BaseException as err:
2184
                raise TypeError(f"Cannot initialize Dataset from {type(data).__name__}") from err
wxchan's avatar
wxchan committed
2185
2186
2187
        if label is not None:
            self.set_label(label)
        if self.get_label() is None:
2188
            raise ValueError("Label should not be None")
wxchan's avatar
wxchan committed
2189
2190
2191
2192
        if weight is not None:
            self.set_weight(weight)
        if group is not None:
            self.set_group(group)
2193
2194
        if position is not None:
            self.set_position(position)
2195
2196
        if isinstance(predictor, _InnerPredictor):
            if self._predictor is None and init_score is not None:
2197
                _log_warning("The init_score will be overridden by the prediction of init_model.")
2198
            self._set_init_score_by_predictor(predictor=predictor, data=data, used_indices=None)
2199
2200
        elif init_score is not None:
            self.set_init_score(init_score)
Guolin Ke's avatar
Guolin Ke committed
2201
        elif predictor is not None:
2202
            raise TypeError(f"Wrong predictor type {type(predictor).__name__}")
Guolin Ke's avatar
Guolin Ke committed
2203
        # set feature names
Nikita Titov's avatar
Nikita Titov committed
2204
        return self.set_feature_name(feature_name)
wxchan's avatar
wxchan committed
2205

2206
    @staticmethod
2207
    def _yield_row_from_seqlist(seqs: List[Sequence], indices: Iterable[int]) -> Iterator[np.ndarray]:
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
        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]
2219
            yield row if row.flags["OWNDATA"] else row.copy()
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232

    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.
2233
        sampled = np.array(list(self._yield_row_from_seqlist(seqs, indices)))
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
        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

2249
2250
2251
    def __init_from_seqs(
        self,
        seqs: List[Sequence],
2252
        ref_dataset: Optional[_DatasetHandle],
2253
    ) -> "Dataset":
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
        """
        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:
2268
            param_str = _param_dict_to_str(self.get_params())
2269
2270
2271
2272
2273
2274
2275
            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)
2276
            batch_size = getattr(seq, "batch_size", None) or Sequence.batch_size
2277
2278
2279
2280
2281
            for start in range(0, nrow, batch_size):
                end = min(start + batch_size, nrow)
                self._push_rows(seq[start:end])
        return self

2282
2283
2284
2285
    def __init_from_np2d(
        self,
        mat: np.ndarray,
        params_str: str,
2286
        ref_dataset: Optional[_DatasetHandle],
2287
    ) -> "Dataset":
2288
        """Initialize data from a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
2289
        if len(mat.shape) != 2:
2290
            raise ValueError("Input numpy.ndarray must be 2 dimensional")
wxchan's avatar
wxchan committed
2291

2292
        self._handle = ctypes.c_void_p()
2293
        data, layout = _np2d_to_np1d(mat)
2294
        ptr_data, type_ptr_data, _ = _c_float_array(data)
2295
2296
2297
2298
2299
2300
        _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]),
2301
                ctypes.c_int(layout),
2302
2303
2304
2305
2306
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2307
        return self
wxchan's avatar
wxchan committed
2308

2309
2310
2311
2312
    def __init_from_list_np2d(
        self,
        mats: List[np.ndarray],
        params_str: str,
2313
        ref_dataset: Optional[_DatasetHandle],
2314
    ) -> "Dataset":
2315
        """Initialize data from a list of 2-D numpy matrices."""
2316
        ncol = mats[0].shape[1]
2317
        nrow = np.empty((len(mats),), np.int32)
2318
        ptr_data: _ctypes_float_array
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))()
2323
        layouts = (ctypes.c_int * len(mats))()
2324
2325

        holders = []
2326
        type_ptr_data = -1
2327
2328
2329

        for i, mat in enumerate(mats):
            if len(mat.shape) != 2:
2330
                raise ValueError("Input numpy.ndarray must be 2 dimensional")
2331
2332

            if mat.shape[1] != ncol:
2333
                raise ValueError("Input arrays must have same number of columns")
2334
2335
2336

            nrow[i] = mat.shape[0]

2337
            mat, layout = _np2d_to_np1d(mat)
2338

2339
            chunk_ptr_data, chunk_type_ptr_data, holder = _c_float_array(mat)
2340
            if type_ptr_data != -1 and chunk_type_ptr_data != type_ptr_data:
2341
                raise ValueError("Input chunks must have same type")
2342
            ptr_data[i] = chunk_ptr_data
2343
            layouts[i] = layout
2344
2345
2346
            type_ptr_data = chunk_type_ptr_data
            holders.append(holder)

2347
        self._handle = ctypes.c_void_p()
2348
2349
2350
2351
2352
2353
2354
        _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),
2355
                layouts,
2356
2357
2358
2359
2360
                _c_str(params_str),
                ref_dataset,
                ctypes.byref(self._handle),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2361
        return self
2362

2363
2364
2365
2366
    def __init_from_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        params_str: str,
2367
        ref_dataset: Optional[_DatasetHandle],
2368
    ) -> "Dataset":
2369
        """Initialize data from a CSR matrix."""
wxchan's avatar
wxchan committed
2370
        if len(csr.indices) != len(csr.data):
2371
            raise ValueError(f"Length mismatch: {len(csr.indices)} vs {len(csr.data)}")
2372
        self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2373

2374
2375
        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
2376

2377
        assert csr.shape[1] <= _MAX_INT32
2378
        csr_indices = csr.indices.astype(np.int32, copy=False)
2379

2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
        _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
2395
        return self
wxchan's avatar
wxchan committed
2396

2397
2398
2399
2400
    def __init_from_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        params_str: str,
2401
        ref_dataset: Optional[_DatasetHandle],
2402
    ) -> "Dataset":
2403
        """Initialize data from a CSC matrix."""
Guolin Ke's avatar
Guolin Ke committed
2404
        if len(csc.indices) != len(csc.data):
2405
            raise ValueError(f"Length mismatch: {len(csc.indices)} vs {len(csc.data)}")
2406
        self._handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
2407

2408
2409
        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
2410

2411
        assert csc.shape[0] <= _MAX_INT32
2412
        csc_indices = csc.indices.astype(np.int32, copy=False)
2413

2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
        _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
2429
        return self
Guolin Ke's avatar
Guolin Ke committed
2430

2431
2432
2433
2434
    def __init_from_pyarrow_table(
        self,
        table: pa_Table,
        params_str: str,
2435
        ref_dataset: Optional[_DatasetHandle],
2436
2437
    ) -> "Dataset":
        """Initialize data from a PyArrow table."""
2438
2439
        if not (PYARROW_INSTALLED and CFFI_INSTALLED):
            raise LightGBMError("Cannot init Dataset from Arrow without 'pyarrow' and 'cffi' installed.")
2440
2441

        # Check that the input is valid: we only handle numbers (for now)
2442
        if not all(arrow_is_integer(t) or arrow_is_floating(t) or arrow_is_boolean(t) for t in table.schema.types):
2443
2444
2445
2446
2447
            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()
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
        _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),
            )
        )
2458
2459
        return self

2460
    @staticmethod
2461
    def _compare_params_for_warning(
2462
2463
        params: Dict[str, Any],
        other_params: Dict[str, Any],
2464
        ignore_keys: Set[str],
2465
2466
    ) -> bool:
        """Compare two dictionaries with params ignoring some keys.
2467

2468
2469
2470
2471
        It is only for the warning purpose.

        Parameters
        ----------
2472
        params : dict
2473
            One dictionary with parameters to compare.
2474
        other_params : dict
2475
2476
2477
            Another dictionary with parameters to compare.
        ignore_keys : set
            Keys that should be ignored during comparing two dictionaries.
2478
2479
2480

        Returns
        -------
2481
2482
        compare_result : bool
          Returns whether two dictionaries with params are equal.
2483
        """
2484
        for k, v in other_params.items():
2485
            if k not in ignore_keys:
2486
                if k not in params or params[k] != v:
2487
                    return False
2488
        for k, v in params.items():
2489
            if k not in ignore_keys:
2490
                if k not in other_params or v != other_params[k]:
2491
2492
2493
                    return False
        return True

2494
    def construct(self) -> "Dataset":
2495
2496
2497
2498
2499
        """Lazy init.

        Returns
        -------
        self : Dataset
Nikita Titov's avatar
Nikita Titov committed
2500
            Constructed Dataset object.
2501
        """
2502
        if self._handle is None:
wxchan's avatar
wxchan committed
2503
            if self.reference is not None:
2504
                reference_params = self.reference.get_params()
2505
2506
                params = self.get_params()
                if params != reference_params:
2507
2508
2509
                    if not self._compare_params_for_warning(
                        params=params,
                        other_params=reference_params,
2510
                        ignore_keys=_ConfigAliases.get("categorical_feature"),
2511
                    ):
2512
                        _log_warning("Overriding the parameters from Reference Dataset.")
2513
                    self._update_params(reference_params)
wxchan's avatar
wxchan committed
2514
                if self.used_indices is None:
2515
                    # create valid
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
                    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
2529
                else:
2530
                    # construct subset
2531
                    used_indices = _list_to_1d_numpy(self.used_indices, dtype=np.int32, name="used_indices")
2532
                    assert used_indices.flags.c_contiguous
Guolin Ke's avatar
Guolin Ke committed
2533
                    if self.reference.group is not None:
2534
                        group_info = np.array(self.reference.group).astype(np.int32, copy=False)
2535
2536
2537
                        _, self.group = np.unique(
                            np.repeat(range(len(group_info)), repeats=group_info)[self.used_indices], return_counts=True
                        )
2538
                    self._handle = ctypes.c_void_p()
2539
                    params_str = _param_dict_to_str(self.params)
2540
2541
2542
2543
2544
2545
2546
2547
2548
                    _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
2549
2550
                    if not self.free_raw_data:
                        self.get_data()
Guolin Ke's avatar
Guolin Ke committed
2551
2552
                    if self.group is not None:
                        self.set_group(self.group)
2553
2554
                    if self.position is not None:
                        self.set_position(self.position)
wxchan's avatar
wxchan committed
2555
2556
                    if self.get_label() is None:
                        raise ValueError("Label should not be None.")
2557
2558
2559
2560
                    if (
                        isinstance(self._predictor, _InnerPredictor)
                        and self._predictor is not self.reference._predictor
                    ):
Guolin Ke's avatar
Guolin Ke committed
2561
                        self.get_data()
2562
                        self._set_init_score_by_predictor(
2563
                            predictor=self._predictor, data=self.data, used_indices=used_indices
2564
                        )
wxchan's avatar
wxchan committed
2565
            else:
2566
                # create train
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
                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
2580
2581
            if self.free_raw_data:
                self.data = None
2582
            self.feature_name = self.get_feature_name()
wxchan's avatar
wxchan committed
2583
        return self
wxchan's avatar
wxchan committed
2584

2585
2586
    def create_valid(
        self,
2587
        data: _LGBM_TrainDataType,
2588
        label: Optional[_LGBM_LabelType] = None,
2589
2590
2591
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
2592
        params: Optional[Dict[str, Any]] = None,
2593
        position: Optional[_LGBM_PositionType] = None,
2594
    ) -> "Dataset":
2595
        """Create validation data align with current Dataset.
wxchan's avatar
wxchan committed
2596
2597
2598

        Parameters
        ----------
2599
        data : str, pathlib.Path, numpy array, pandas DataFrame, scipy.sparse, Sequence, list of Sequence, list of numpy array or pyarrow Table
wxchan's avatar
wxchan committed
2600
            Data source of Dataset.
2601
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
2602
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2603
            Label of the data.
2604
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2605
            Weight for each instance. Weights should be non-negative.
2606
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2607
2608
2609
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2610
2611
            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.
2612
        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)
2613
            Init score for Dataset.
Nikita Titov's avatar
Nikita Titov committed
2614
        params : dict or None, optional (default=None)
2615
            Other parameters for validation Dataset.
2616
2617
        position : numpy 1-D array, pandas Series or None, optional (default=None)
            Position of items used in unbiased learning-to-rank task.
2618
2619
2620

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2621
2622
        valid : Dataset
            Validation Dataset with reference to self.
wxchan's avatar
wxchan committed
2623
        """
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
        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
2635
        ret._predictor = self._predictor
2636
        ret.pandas_categorical = self.pandas_categorical
wxchan's avatar
wxchan committed
2637
        return ret
wxchan's avatar
wxchan committed
2638

2639
2640
2641
    def subset(
        self,
        used_indices: List[int],
2642
        params: Optional[Dict[str, Any]] = None,
2643
    ) -> "Dataset":
2644
        """Get subset of current Dataset.
wxchan's avatar
wxchan committed
2645
2646
2647
2648

        Parameters
        ----------
        used_indices : list of int
2649
            Indices used to create the subset.
Nikita Titov's avatar
Nikita Titov committed
2650
        params : dict or None, optional (default=None)
2651
            These parameters will be passed to Dataset constructor.
2652
2653
2654
2655
2656

        Returns
        -------
        subset : Dataset
            Subset of the current Dataset.
wxchan's avatar
wxchan committed
2657
        """
wxchan's avatar
wxchan committed
2658
2659
        if params is None:
            params = self.params
2660
2661
2662
2663
2664
2665
2666
2667
        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
2668
        ret._predictor = self._predictor
2669
        ret.pandas_categorical = self.pandas_categorical
2670
        ret.used_indices = sorted(used_indices)
wxchan's avatar
wxchan committed
2671
2672
        return ret

2673
    def save_binary(self, filename: Union[str, Path]) -> "Dataset":
2674
        """Save Dataset to a binary file.
wxchan's avatar
wxchan committed
2675

2676
2677
2678
2679
2680
        .. 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
2681
2682
        Parameters
        ----------
2683
        filename : str or pathlib.Path
wxchan's avatar
wxchan committed
2684
            Name of the output file.
Nikita Titov's avatar
Nikita Titov committed
2685
2686
2687
2688
2689

        Returns
        -------
        self : Dataset
            Returns self.
wxchan's avatar
wxchan committed
2690
        """
2691
2692
2693
2694
2695
2696
        _safe_call(
            _LIB.LGBM_DatasetSaveBinary(
                self.construct()._handle,
                _c_str(str(filename)),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
2697
        return self
wxchan's avatar
wxchan committed
2698

2699
    def _update_params(self, params: Optional[Dict[str, Any]]) -> "Dataset":
2700
2701
        if not params:
            return self
2702
        params = deepcopy(params)
2703

2704
        def update() -> None:
2705
2706
2707
            if not self.params:
                self.params = params
            else:
2708
                self._params_back_up = deepcopy(self.params)
2709
2710
                self.params.update(params)

2711
        if self._handle is None:
2712
2713
2714
            update()
        elif params is not None:
            ret = _LIB.LGBM_DatasetUpdateParamChecking(
2715
                _c_str(_param_dict_to_str(self.params)),
2716
2717
                _c_str(_param_dict_to_str(params)),
            )
2718
2719
2720
2721
2722
2723
            if ret != 0:
                # could be updated if data is not freed
                if self.data is not None:
                    update()
                    self._free_handle()
                else:
2724
                    raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
Nikita Titov's avatar
Nikita Titov committed
2725
        return self
wxchan's avatar
wxchan committed
2726

2727
    def _reverse_update_params(self) -> "Dataset":
2728
        if self._handle is None:
2729
2730
            self.params = deepcopy(self._params_back_up)
            self._params_back_up = None
Nikita Titov's avatar
Nikita Titov committed
2731
        return self
2732

2733
2734
2735
    def set_field(
        self,
        field_name: str,
2736
        data: Optional[_LGBM_SetFieldType],
2737
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2738
        """Set property into the Dataset.
wxchan's avatar
wxchan committed
2739
2740
2741

        Parameters
        ----------
2742
        field_name : str
2743
            The field name of the information.
2744
        data : 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
2745
            The data to be set.
Nikita Titov's avatar
Nikita Titov committed
2746
2747
2748
2749
2750

        Returns
        -------
        self : Dataset
            Dataset with set property.
wxchan's avatar
wxchan committed
2751
        """
2752
        if self._handle is None:
2753
            raise Exception(f"Cannot set {field_name} before construct dataset")
wxchan's avatar
wxchan committed
2754
        if data is None:
2755
            # set to None
2756
2757
2758
2759
2760
2761
2762
2763
2764
            _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
2765
            return self
2766
2767

        # If the data is a arrow data, we can just pass it to C
2768
2769
2770
2771
2772
2773
        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}'")
2774
2775
2776
2777
2778
2779
2780
                data = pa_chunked_array(
                    [
                        chunk
                        for array in data.columns  # type: ignore
                        for chunk in array.chunks
                    ]
                )
2781

2782
            c_array = _export_arrow_to_c(data)
2783
2784
2785
2786
2787
2788
2789
2790
2791
            _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),
                )
            )
2792
2793
2794
            self.version += 1
            return self

2795
        dtype: "np.typing.DTypeLike"
2796
        if field_name == "init_score":
Guolin Ke's avatar
Guolin Ke committed
2797
            dtype = np.float64
2798
            if _is_1d_collection(data):
2799
                data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2800
            elif _is_2d_collection(data):
2801
                data = _data_to_2d_numpy(data, dtype=dtype, name=field_name)
2802
                data = data.ravel(order="F")
2803
2804
            else:
                raise TypeError(
2805
2806
                    "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."
2807
2808
                )
        else:
2809
            if field_name in {"group", "position"}:
2810
2811
2812
                dtype = np.int32
            else:
                dtype = np.float32
2813
            data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2814

2815
        ptr_data: Union[_ctypes_float_ptr, _ctypes_int_ptr]
2816
        if data.dtype == np.float32 or data.dtype == np.float64:
2817
            ptr_data, type_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
2818
        elif data.dtype == np.int32:
2819
            ptr_data, type_data, _ = _c_int_array(data)
wxchan's avatar
wxchan committed
2820
        else:
2821
            raise TypeError(f"Expected np.float32/64 or np.int32, met type({data.dtype})")
2822
        if type_data != _FIELD_TYPE_MAPPER[field_name]:
2823
            raise TypeError("Input type error for set_field")
2824
2825
2826
2827
2828
2829
2830
2831
2832
        _safe_call(
            _LIB.LGBM_DatasetSetField(
                self._handle,
                _c_str(field_name),
                ptr_data,
                ctypes.c_int(len(data)),
                ctypes.c_int(type_data),
            )
        )
2833
        self.version += 1
Nikita Titov's avatar
Nikita Titov committed
2834
        return self
wxchan's avatar
wxchan committed
2835

2836
    def get_field(self, field_name: str) -> Optional[np.ndarray]:
wxchan's avatar
wxchan committed
2837
        """Get property from the Dataset.
wxchan's avatar
wxchan committed
2838

2839
2840
2841
2842
2843
2844
        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
2845
2846
        Parameters
        ----------
2847
        field_name : str
2848
            The field name of the information.
wxchan's avatar
wxchan committed
2849
2850
2851

        Returns
        -------
2852
        info : numpy array or None
2853
            A numpy array with information from the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2854
        """
2855
        if self._handle is None:
2856
            raise Exception(f"Cannot get {field_name} before construct Dataset")
2857
2858
        tmp_out_len = ctypes.c_int(0)
        out_type = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2859
        ret = ctypes.POINTER(ctypes.c_void_p)()
2860
2861
2862
2863
2864
2865
2866
2867
2868
        _safe_call(
            _LIB.LGBM_DatasetGetField(
                self._handle,
                _c_str(field_name),
                ctypes.byref(tmp_out_len),
                ctypes.byref(ret),
                ctypes.byref(out_type),
            )
        )
2869
        if out_type.value != _FIELD_TYPE_MAPPER[field_name]:
wxchan's avatar
wxchan committed
2870
2871
2872
            raise TypeError("Return type error for get_field")
        if tmp_out_len.value == 0:
            return None
2873
        if out_type.value == _C_API_DTYPE_INT32:
2874
2875
            arr = _cint32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_int32)),
2876
                length=tmp_out_len.value,
2877
            )
2878
        elif out_type.value == _C_API_DTYPE_FLOAT32:
2879
2880
            arr = _cfloat32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_float)),
2881
                length=tmp_out_len.value,
2882
            )
2883
        elif out_type.value == _C_API_DTYPE_FLOAT64:
2884
2885
            arr = _cfloat64_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_double)),
2886
                length=tmp_out_len.value,
2887
            )
2888
        else:
wxchan's avatar
wxchan committed
2889
            raise TypeError("Unknown type")
2890
        if field_name == "init_score":
2891
2892
2893
            num_data = self.num_data()
            num_classes = arr.size // num_data
            if num_classes > 1:
2894
                arr = arr.reshape((num_data, num_classes), order="F")
2895
        return arr
Guolin Ke's avatar
Guolin Ke committed
2896

2897
2898
    def set_categorical_feature(
        self,
2899
        categorical_feature: _LGBM_CategoricalFeatureConfiguration,
2900
    ) -> "Dataset":
2901
        """Set categorical features.
2902
2903
2904

        Parameters
        ----------
2905
        categorical_feature : list of str or int, or 'auto'
2906
            Names or indices of categorical features.
Nikita Titov's avatar
Nikita Titov committed
2907
2908
2909
2910
2911

        Returns
        -------
        self : Dataset
            Dataset with set categorical features.
2912
2913
        """
        if self.categorical_feature == categorical_feature:
Nikita Titov's avatar
Nikita Titov committed
2914
            return self
2915
        if self.data is not None:
2916
2917
            if self.categorical_feature is None:
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2918
                return self._free_handle()
2919
            elif categorical_feature == "auto":
Nikita Titov's avatar
Nikita Titov committed
2920
                return self
2921
            else:
2922
2923
2924
2925
2926
                if self.categorical_feature != "auto":
                    _log_warning(
                        "categorical_feature in Dataset is overridden.\n"
                        f"New categorical_feature is {list(categorical_feature)}"
                    )
2927
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2928
                return self._free_handle()
2929
        else:
2930
2931
2932
2933
            raise LightGBMError(
                "Cannot set categorical feature after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
2934

2935
2936
    def _set_predictor(
        self,
2937
        predictor: Optional[_InnerPredictor],
2938
    ) -> "Dataset":
2939
2940
2941
2942
        """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
2943
        """
2944
        if predictor is None and self._predictor is None:
Nikita Titov's avatar
Nikita Titov committed
2945
            return self
2946
        elif isinstance(predictor, _InnerPredictor) and isinstance(self._predictor, _InnerPredictor):
2947
2948
2949
            if (predictor == self._predictor) and (
                predictor.current_iteration() == self._predictor.current_iteration()
            ):
2950
                return self
2951
        if self._handle is None:
Guolin Ke's avatar
Guolin Ke committed
2952
            self._predictor = predictor
2953
2954
        elif self.data is not None:
            self._predictor = predictor
2955
2956
2957
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.data,
2958
                used_indices=None,
2959
            )
2960
2961
        elif self.used_indices is not None and self.reference is not None and self.reference.data is not None:
            self._predictor = predictor
2962
2963
2964
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.reference.data,
2965
                used_indices=self.used_indices,
2966
            )
Guolin Ke's avatar
Guolin Ke committed
2967
        else:
2968
2969
2970
2971
            raise LightGBMError(
                "Cannot set predictor after freed raw data, "
                "set free_raw_data=False when construct Dataset to avoid this."
            )
2972
        return self
Guolin Ke's avatar
Guolin Ke committed
2973

2974
    def set_reference(self, reference: "Dataset") -> "Dataset":
2975
        """Set reference Dataset.
Guolin Ke's avatar
Guolin Ke committed
2976
2977
2978
2979

        Parameters
        ----------
        reference : Dataset
2980
            Reference that is used as a template to construct the current Dataset.
Nikita Titov's avatar
Nikita Titov committed
2981
2982
2983
2984
2985

        Returns
        -------
        self : Dataset
            Dataset with set reference.
Guolin Ke's avatar
Guolin Ke committed
2986
        """
2987
2988
2989
        self.set_categorical_feature(reference.categorical_feature).set_feature_name(
            reference.feature_name
        )._set_predictor(reference._predictor)
2990
        # we're done if self and reference share a common upstream reference
2991
        if self.get_ref_chain().intersection(reference.get_ref_chain()):
Nikita Titov's avatar
Nikita Titov committed
2992
            return self
Guolin Ke's avatar
Guolin Ke committed
2993
2994
        if self.data is not None:
            self.reference = reference
Nikita Titov's avatar
Nikita Titov committed
2995
            return self._free_handle()
Guolin Ke's avatar
Guolin Ke committed
2996
        else:
2997
2998
2999
3000
            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
3001

3002
    def set_feature_name(self, feature_name: _LGBM_FeatureNameConfiguration) -> "Dataset":
3003
        """Set feature name.
Guolin Ke's avatar
Guolin Ke committed
3004
3005
3006

        Parameters
        ----------
3007
        feature_name : list of str
3008
            Feature names.
Nikita Titov's avatar
Nikita Titov committed
3009
3010
3011
3012
3013

        Returns
        -------
        self : Dataset
            Dataset with set feature name.
Guolin Ke's avatar
Guolin Ke committed
3014
        """
3015
        if feature_name != "auto":
3016
            self.feature_name = feature_name
3017
        if self._handle is not None and feature_name is not None and feature_name != "auto":
wxchan's avatar
wxchan committed
3018
            if len(feature_name) != self.num_feature():
3019
3020
3021
                raise ValueError(
                    f"Length of feature_name({len(feature_name)}) and num_feature({self.num_feature()}) don't match"
                )
3022
            c_feature_name = [_c_str(name) for name in feature_name]
3023
3024
3025
3026
3027
3028
3029
            _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
3030
        return self
Guolin Ke's avatar
Guolin Ke committed
3031

3032
    def set_label(self, label: Optional[_LGBM_LabelType]) -> "Dataset":
3033
        """Set label of Dataset.
Guolin Ke's avatar
Guolin Ke committed
3034
3035
3036

        Parameters
        ----------
3037
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None
3038
            The label information to be set into Dataset.
Nikita Titov's avatar
Nikita Titov committed
3039
3040
3041
3042
3043

        Returns
        -------
        self : Dataset
            Dataset with set label.
Guolin Ke's avatar
Guolin Ke committed
3044
3045
        """
        self.label = label
3046
        if self._handle is not None:
3047
3048
            if isinstance(label, pd_DataFrame):
                if len(label.columns) > 1:
3049
                    raise ValueError("DataFrame for label cannot have multiple columns")
3050
                label_array = np.ravel(_pandas_to_numpy(label, target_dtype=np.float32))
3051
3052
            elif _is_pyarrow_array(label):
                label_array = label
3053
            else:
3054
3055
3056
                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
3057
        return self
Guolin Ke's avatar
Guolin Ke committed
3058

3059
3060
    def set_weight(
        self,
3061
        weight: Optional[_LGBM_WeightType],
3062
    ) -> "Dataset":
3063
        """Set weight of each instance.
Guolin Ke's avatar
Guolin Ke committed
3064
3065
3066

        Parameters
        ----------
3067
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3068
            Weight to be set for each data point. Weights should be non-negative.
Nikita Titov's avatar
Nikita Titov committed
3069
3070
3071
3072
3073

        Returns
        -------
        self : Dataset
            Dataset with set weight.
Guolin Ke's avatar
Guolin Ke committed
3074
        """
3075
3076
3077
3078
3079
3080
3081
        # 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
3082
        self.weight = weight
3083
3084

        # Set field
3085
        if self._handle is not None and weight is not None:
3086
            if not _is_pyarrow_array(weight):
3087
3088
3089
                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
3090
        return self
Guolin Ke's avatar
Guolin Ke committed
3091

3092
3093
    def set_init_score(
        self,
3094
        init_score: Optional[_LGBM_InitScoreType],
3095
    ) -> "Dataset":
3096
        """Set init score of Booster to start from.
Guolin Ke's avatar
Guolin Ke committed
3097
3098
3099

        Parameters
        ----------
3100
        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
3101
            Init score for Booster.
Nikita Titov's avatar
Nikita Titov committed
3102
3103
3104
3105
3106

        Returns
        -------
        self : Dataset
            Dataset with set init score.
Guolin Ke's avatar
Guolin Ke committed
3107
3108
        """
        self.init_score = init_score
3109
        if self._handle is not None and init_score is not None:
3110
3111
            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
3112
        return self
Guolin Ke's avatar
Guolin Ke committed
3113

3114
3115
    def set_group(
        self,
3116
        group: Optional[_LGBM_GroupType],
3117
    ) -> "Dataset":
3118
        """Set group size of Dataset (used for ranking).
Guolin Ke's avatar
Guolin Ke committed
3119
3120
3121

        Parameters
        ----------
3122
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3123
3124
3125
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
3126
3127
            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
3128
3129
3130
3131
3132

        Returns
        -------
        self : Dataset
            Dataset with set group.
Guolin Ke's avatar
Guolin Ke committed
3133
3134
        """
        self.group = group
3135
        if self._handle is not None and group is not None:
3136
            if not _is_pyarrow_array(group):
3137
3138
                group = _list_to_1d_numpy(group, dtype=np.int32, name="group")
            self.set_field("group", group)
3139
            # original values can be modified at cpp side
3140
            constructed_group = self.get_field("group")
3141
3142
            if constructed_group is not None:
                self.group = np.diff(constructed_group)
Nikita Titov's avatar
Nikita Titov committed
3143
        return self
Guolin Ke's avatar
Guolin Ke committed
3144

3145
3146
    def set_position(
        self,
3147
        position: Optional[_LGBM_PositionType],
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
    ) -> "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:
3163
3164
            position = _list_to_1d_numpy(position, dtype=np.int32, name="position")
            self.set_field("position", position)
3165
3166
        return self

3167
    def get_feature_name(self) -> List[str]:
3168
3169
3170
3171
        """Get the names of columns (features) in the Dataset.

        Returns
        -------
3172
        feature_names : list of str
3173
3174
            The names of columns (features) in the Dataset.
        """
3175
        if self._handle is None:
3176
3177
3178
3179
3180
            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)
3181
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
3182
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
        _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,
            )
        )
3193
3194
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
3195
3196
3197
3198
        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)]
3199
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
            _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)]
3211

3212
    def get_label(self) -> Optional[_LGBM_LabelType]:
3213
        """Get the label of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3214
3215
3216

        Returns
        -------
3217
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None
3218
            The label information from the Dataset.
3219
            For a constructed ``Dataset``, this will only return a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3220
        """
3221
        if self.label is None:
3222
            self.label = self.get_field("label")
Guolin Ke's avatar
Guolin Ke committed
3223
3224
        return self.label

3225
    def get_weight(self) -> Optional[_LGBM_WeightType]:
3226
        """Get the weight of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3227
3228
3229

        Returns
        -------
3230
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3231
            Weight for each data point from the Dataset. Weights should be non-negative.
3232
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3233
        """
3234
        if self.weight is None:
3235
            self.weight = self.get_field("weight")
Guolin Ke's avatar
Guolin Ke committed
3236
3237
        return self.weight

3238
    def get_init_score(self) -> Optional[_LGBM_InitScoreType]:
3239
        """Get the initial score of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3240
3241
3242

        Returns
        -------
3243
        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
3244
            Init score of Booster.
3245
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3246
        """
3247
        if self.init_score is None:
3248
            self.init_score = self.get_field("init_score")
Guolin Ke's avatar
Guolin Ke committed
3249
3250
        return self.init_score

3251
    def get_data(self) -> Optional[_LGBM_TrainDataType]:
3252
3253
3254
3255
        """Get the raw data of the Dataset.

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

3284
    def get_group(self) -> Optional[_LGBM_GroupType]:
3285
        """Get the group of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3286
3287
3288

        Returns
        -------
3289
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
3290
3291
3292
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
3293
3294
            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.
3295
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3296
        """
3297
        if self.group is None:
3298
            self.group = self.get_field("group")
Guolin Ke's avatar
Guolin Ke committed
3299
3300
            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
3301
                self.group = np.diff(self.group)
Guolin Ke's avatar
Guolin Ke committed
3302
3303
        return self.group

3304
    def get_position(self) -> Optional[_LGBM_PositionType]:
3305
3306
3307
3308
        """Get the position of the Dataset.

        Returns
        -------
3309
        position : numpy 1-D array, pandas Series or None
3310
            Position of items used in unbiased learning-to-rank task.
3311
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
3312
3313
        """
        if self.position is None:
3314
            self.position = self.get_field("position")
3315
3316
        return self.position

3317
    def num_data(self) -> int:
3318
        """Get the number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3319
3320
3321

        Returns
        -------
3322
3323
        number_of_rows : int
            The number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3324
        """
3325
        if self._handle is not None:
3326
            ret = ctypes.c_int(0)
3327
3328
3329
3330
3331
3332
            _safe_call(
                _LIB.LGBM_DatasetGetNumData(
                    self._handle,
                    ctypes.byref(ret),
                )
            )
wxchan's avatar
wxchan committed
3333
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3334
        else:
3335
            raise LightGBMError("Cannot get num_data before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3336

3337
    def num_feature(self) -> int:
3338
        """Get the number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3339
3340
3341

        Returns
        -------
3342
3343
        number_of_columns : int
            The number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3344
        """
3345
        if self._handle is not None:
3346
            ret = ctypes.c_int(0)
3347
3348
3349
3350
3351
3352
            _safe_call(
                _LIB.LGBM_DatasetGetNumFeature(
                    self._handle,
                    ctypes.byref(ret),
                )
            )
wxchan's avatar
wxchan committed
3353
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3354
        else:
3355
            raise LightGBMError("Cannot get num_feature before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3356

3357
    def feature_num_bin(self, feature: Union[int, str]) -> int:
3358
3359
        """Get the number of bins for a feature.

3360
3361
        .. versionadded:: 4.0.0

3362
3363
        Parameters
        ----------
3364
3365
        feature : int or str
            Index or name of the feature.
3366
3367
3368
3369
3370
3371

        Returns
        -------
        number_of_bins : int
            The number of constructed bins for the feature in the Dataset.
        """
3372
        if self._handle is not None:
3373
            if isinstance(feature, str):
3374
3375
3376
                feature_index = self.feature_name.index(feature)
            else:
                feature_index = feature
3377
            ret = ctypes.c_int(0)
3378
3379
3380
3381
3382
3383
3384
            _safe_call(
                _LIB.LGBM_DatasetGetFeatureNumBin(
                    self._handle,
                    ctypes.c_int(feature_index),
                    ctypes.byref(ret),
                )
            )
3385
3386
3387
3388
            return ret.value
        else:
            raise LightGBMError("Cannot get feature_num_bin before construct dataset")

3389
    def get_ref_chain(self, ref_limit: int = 100) -> Set["Dataset"]:
3390
3391
3392
3393
3394
        """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.
3395
3396
3397
3398
3399

        Parameters
        ----------
        ref_limit : int, optional (default=100)
            The limit number of references.
3400
3401
3402

        Returns
        -------
3403
3404
3405
        ref_chain : set of Dataset
            Chain of references of the Datasets.
        """
3406
        head = self
3407
        ref_chain: Set[Dataset] = set()
3408
3409
        while len(ref_chain) < ref_limit:
            if isinstance(head, Dataset):
3410
                ref_chain.add(head)
3411
3412
3413
3414
3415
3416
                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
3417
        return ref_chain
3418

3419
    def add_features_from(self, other: "Dataset") -> "Dataset":
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
        """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.
        """
3434
        if self._handle is None or other._handle is None:
3435
3436
3437
3438
3439
3440
3441
            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
3442
3443
3444
3445
3446
3447
3448
3449
        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))
3450
                elif isinstance(other.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3451
                    self.data = np.hstack((self.data, other.data.toarray()))
3452
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3453
3454
3455
                    self.data = np.hstack((self.data, other.data.values))
                else:
                    self.data = None
3456
            elif isinstance(self.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3457
                sparse_format = self.data.getformat()
3458
                if isinstance(other.data, (np.ndarray, scipy.sparse.spmatrix)):
Guolin Ke's avatar
Guolin Ke committed
3459
                    self.data = scipy.sparse.hstack((self.data, other.data), format=sparse_format)
3460
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3461
3462
3463
                    self.data = scipy.sparse.hstack((self.data, other.data.values), format=sparse_format)
                else:
                    self.data = None
3464
            elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3465
                if not PANDAS_INSTALLED:
3466
3467
3468
3469
3470
                    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
3471
                if isinstance(other.data, np.ndarray):
3472
                    self.data = concat((self.data, pd_DataFrame(other.data)), axis=1, ignore_index=True)
3473
                elif isinstance(other.data, scipy.sparse.spmatrix):
3474
                    self.data = concat((self.data, pd_DataFrame(other.data.toarray())), axis=1, ignore_index=True)
3475
                elif isinstance(other.data, pd_DataFrame):
3476
                    self.data = concat((self.data, other.data), axis=1, ignore_index=True)
Guolin Ke's avatar
Guolin Ke committed
3477
3478
3479
3480
3481
                else:
                    self.data = None
            else:
                self.data = None
        if self.data is None:
3482
3483
3484
3485
3486
3487
3488
            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"
            )
3489
            _log_warning(err_msg)
Guolin Ke's avatar
Guolin Ke committed
3490
        self.feature_name = self.get_feature_name()
3491
        _log_warning(
3492
            "Resetting categorical features.\n"
3493
3494
            "You can set new categorical features via ``set_categorical_feature`` method"
        )
Guolin Ke's avatar
Guolin Ke committed
3495
3496
        self.categorical_feature = "auto"
        self.pandas_categorical = None
3497
3498
        return self

3499
    def _dump_text(self, filename: Union[str, Path]) -> "Dataset":
3500
3501
3502
3503
3504
3505
        """Save Dataset to a text file.

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

        Parameters
        ----------
3506
        filename : str or pathlib.Path
3507
3508
3509
3510
3511
3512
3513
            Name of the output file.

        Returns
        -------
        self : Dataset
            Returns self.
        """
3514
3515
3516
3517
3518
3519
        _safe_call(
            _LIB.LGBM_DatasetDumpText(
                self.construct()._handle,
                _c_str(str(filename)),
            )
        )
3520
3521
        return self

wxchan's avatar
wxchan committed
3522

3523
3524
_LGBM_CustomObjectiveFunction = Callable[
    [np.ndarray, Dataset],
3525
    Tuple[np.ndarray, np.ndarray],
3526
]
3527
3528
3529
_LGBM_CustomEvalFunction = Union[
    Callable[
        [np.ndarray, Dataset],
3530
        _LGBM_EvalFunctionResultType,
3531
3532
3533
    ],
    Callable[
        [np.ndarray, Dataset],
3534
3535
        List[_LGBM_EvalFunctionResultType],
    ],
3536
]
3537
3538


3539
class Booster:
3540
    """Booster in LightGBM."""
3541

3542
3543
3544
3545
3546
    def __init__(
        self,
        params: Optional[Dict[str, Any]] = None,
        train_set: Optional[Dataset] = None,
        model_file: Optional[Union[str, Path]] = None,
3547
        model_str: Optional[str] = None,
3548
    ):
3549
        """Initialize the Booster.
wxchan's avatar
wxchan committed
3550
3551
3552

        Parameters
        ----------
Nikita Titov's avatar
Nikita Titov committed
3553
        params : dict or None, optional (default=None)
3554
3555
3556
            Parameters for Booster.
        train_set : Dataset or None, optional (default=None)
            Training dataset.
3557
        model_file : str, pathlib.Path or None, optional (default=None)
wxchan's avatar
wxchan committed
3558
            Path to the model file.
3559
        model_str : str or None, optional (default=None)
3560
            Model will be loaded from this string.
wxchan's avatar
wxchan committed
3561
        """
3562
        self._handle = ctypes.c_void_p()
3563
        self._network = False
wxchan's avatar
wxchan committed
3564
        self.__need_reload_eval_info = True
3565
        self._train_data_name = "training"
3566
        self.__set_objective_to_none = False
wxchan's avatar
wxchan committed
3567
        self.best_iteration = -1
3568
        self.best_score: _LGBM_BoosterBestScoreType = {}
3569
        params = {} if params is None else deepcopy(params)
wxchan's avatar
wxchan committed
3570
        if train_set is not None:
3571
            # Training task
wxchan's avatar
wxchan committed
3572
            if not isinstance(train_set, Dataset):
3573
                raise TypeError(f"Training data should be Dataset instance, met {type(train_set).__name__}")
3574
3575
3576
            params = _choose_param_value(
                main_param_name="machines",
                params=params,
3577
                default_value=None,
3578
3579
3580
3581
3582
3583
3584
            )
            # 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):
3585
                    num_machines_from_machine_list = len(machines.split(","))
3586
3587
                elif isinstance(machines, (list, set)):
                    num_machines_from_machine_list = len(machines)
3588
                    machines = ",".join(machines)
3589
3590
3591
3592
3593
3594
                else:
                    raise ValueError("Invalid machines in params.")

                params = _choose_param_value(
                    main_param_name="num_machines",
                    params=params,
3595
                    default_value=num_machines_from_machine_list,
3596
3597
3598
3599
                )
                params = _choose_param_value(
                    main_param_name="local_listen_port",
                    params=params,
3600
                    default_value=12400,
3601
3602
3603
3604
3605
                )
                self.set_network(
                    machines=machines,
                    local_listen_port=params["local_listen_port"],
                    listen_time_out=params.get("time_out", 120),
3606
                    num_machines=params["num_machines"],
3607
                )
3608
            # construct booster object
3609
3610
3611
            train_set.construct()
            # copy the parameters from train_set
            params.update(train_set.get_params())
3612
            params_str = _param_dict_to_str(params)
3613
3614
3615
3616
3617
3618
3619
            _safe_call(
                _LIB.LGBM_BoosterCreate(
                    train_set._handle,
                    _c_str(params_str),
                    ctypes.byref(self._handle),
                )
            )
3620
            # save reference to data
wxchan's avatar
wxchan committed
3621
            self.train_set = train_set
3622
3623
            self.valid_sets: List[Dataset] = []
            self.name_valid_sets: List[str] = []
wxchan's avatar
wxchan committed
3624
            self.__num_dataset = 1
Guolin Ke's avatar
Guolin Ke committed
3625
3626
            self.__init_predictor = train_set._predictor
            if self.__init_predictor is not None:
3627
3628
3629
3630
3631
3632
                _safe_call(
                    _LIB.LGBM_BoosterMerge(
                        self._handle,
                        self.__init_predictor._handle,
                    )
                )
Guolin Ke's avatar
Guolin Ke committed
3633
            out_num_class = ctypes.c_int(0)
3634
3635
3636
3637
3638
3639
            _safe_call(
                _LIB.LGBM_BoosterGetNumClasses(
                    self._handle,
                    ctypes.byref(out_num_class),
                )
            )
wxchan's avatar
wxchan committed
3640
            self.__num_class = out_num_class.value
3641
            # buffer for inner predict
3642
            self.__inner_predict_buffer: List[Optional[np.ndarray]] = [None]
wxchan's avatar
wxchan committed
3643
3644
            self.__is_predicted_cur_iter = [False]
            self.__get_eval_info()
3645
            self.pandas_categorical = train_set.pandas_categorical
3646
            self.train_set_version = train_set.version
wxchan's avatar
wxchan committed
3647
        elif model_file is not None:
3648
            # Prediction task
Guolin Ke's avatar
Guolin Ke committed
3649
            out_num_iterations = ctypes.c_int(0)
3650
3651
3652
3653
3654
3655
3656
            _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
3657
            out_num_class = ctypes.c_int(0)
3658
3659
3660
3661
3662
3663
            _safe_call(
                _LIB.LGBM_BoosterGetNumClasses(
                    self._handle,
                    ctypes.byref(out_num_class),
                )
            )
wxchan's avatar
wxchan committed
3664
            self.__num_class = out_num_class.value
3665
            self.pandas_categorical = _load_pandas_categorical(file_name=model_file)
3666
            if params:
3667
                _log_warning("Ignoring params argument, using parameters from model file.")
3668
            params = self._get_loaded_param()
3669
        elif model_str is not None:
3670
            self.model_from_string(model_str)
3671
3672
3673
            if params:
                _log_warning("Ignoring params argument, using parameters from model string.")
            params = self._get_loaded_param()
wxchan's avatar
wxchan committed
3674
        else:
3675
            raise TypeError(
3676
                "Need at least one training dataset or model file or model string to create Booster instance"
3677
            )
3678
        self.params = params
wxchan's avatar
wxchan committed
3679

3680
    def __del__(self) -> None:
3681
        try:
3682
            if self._network:
3683
3684
3685
3686
                self.free_network()
        except AttributeError:
            pass
        try:
3687
3688
            if self._handle is not None:
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
3689
3690
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
3691

3692
    def __copy__(self) -> "Booster":
wxchan's avatar
wxchan committed
3693
3694
        return self.__deepcopy__(None)

3695
    def __deepcopy__(self, *args: Any, **kwargs: Any) -> "Booster":
3696
        model_str = self.model_to_string(num_iteration=-1)
3697
        return Booster(model_str=model_str)
wxchan's avatar
wxchan committed
3698

3699
    def __getstate__(self) -> Dict[str, Any]:
wxchan's avatar
wxchan committed
3700
        this = self.__dict__.copy()
3701
3702
3703
        handle = this["_handle"]
        this.pop("train_set", None)
        this.pop("valid_sets", None)
wxchan's avatar
wxchan committed
3704
        if handle is not None:
3705
            this["_handle"] = self.model_to_string(num_iteration=-1)
wxchan's avatar
wxchan committed
3706
3707
        return this

3708
    def __setstate__(self, state: Dict[str, Any]) -> None:
3709
        model_str = state.get("_handle", state.get("handle", None))
3710
        if model_str is not None:
wxchan's avatar
wxchan committed
3711
            handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
3712
            out_num_iterations = ctypes.c_int(0)
3713
3714
3715
3716
3717
3718
3719
3720
            _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
3721
3722
        self.__dict__.update(state)

3723
3724
3725
3726
    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)
3727
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3728
3729
3730
3731
3732
3733
3734
3735
        _safe_call(
            _LIB.LGBM_BoosterGetLoadedParam(
                self._handle,
                ctypes.c_int64(buffer_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer,
            )
        )
3736
3737
3738
3739
        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)
3740
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3741
3742
3743
3744
3745
3746
3747
3748
3749
            _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"))
3750

3751
    def free_dataset(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3752
3753
3754
3755
3756
3757
3758
        """Free Booster's Datasets.

        Returns
        -------
        self : Booster
            Booster without Datasets.
        """
3759
3760
        self.__dict__.pop("train_set", None)
        self.__dict__.pop("valid_sets", None)
3761
        self.__num_dataset = 0
Nikita Titov's avatar
Nikita Titov committed
3762
        return self
wxchan's avatar
wxchan committed
3763

3764
    def _free_buffer(self) -> "Booster":
3765
3766
        self.__inner_predict_buffer = []
        self.__is_predicted_cur_iter = []
Nikita Titov's avatar
Nikita Titov committed
3767
        return self
3768

3769
3770
3771
3772
3773
    def set_network(
        self,
        machines: Union[List[str], Set[str], str],
        local_listen_port: int = 12400,
        listen_time_out: int = 120,
3774
        num_machines: int = 1,
3775
    ) -> "Booster":
3776
3777
3778
3779
        """Set the network configuration.

        Parameters
        ----------
3780
        machines : list, set or str
3781
            Names of machines.
Nikita Titov's avatar
Nikita Titov committed
3782
        local_listen_port : int, optional (default=12400)
3783
            TCP listen port for local machines.
Nikita Titov's avatar
Nikita Titov committed
3784
        listen_time_out : int, optional (default=120)
3785
            Socket time-out in minutes.
Nikita Titov's avatar
Nikita Titov committed
3786
        num_machines : int, optional (default=1)
3787
            The number of machines for distributed learning application.
Nikita Titov's avatar
Nikita Titov committed
3788
3789
3790
3791
3792

        Returns
        -------
        self : Booster
            Booster with set network.
3793
        """
3794
        if isinstance(machines, (list, set)):
3795
3796
3797
3798
3799
3800
3801
3802
3803
            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),
            )
        )
3804
        self._network = True
Nikita Titov's avatar
Nikita Titov committed
3805
        return self
3806

3807
    def free_network(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3808
3809
3810
3811
3812
3813
3814
        """Free Booster's network.

        Returns
        -------
        self : Booster
            Booster with freed network.
        """
3815
        _safe_call(_LIB.LGBM_NetworkFree())
3816
        self._network = False
Nikita Titov's avatar
Nikita Titov committed
3817
        return self
3818

3819
    def trees_to_dataframe(self) -> pd_DataFrame:
3820
3821
        """Parse the fitted model and return in an easy-to-read pandas DataFrame.

3822
3823
3824
3825
        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.
3826
3827
3828
3829
3830
            - ``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.
3831
3832
            - ``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.
3833
            - ``decision_type`` : str, logical operator describing how to compare a value to ``threshold``.
3834
3835
              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.
3836
3837
            - ``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.
3838
            - ``value`` : float64, predicted value for this leaf node, multiplied by the learning rate.
3839
            - ``weight`` : float64 or int64, sum of Hessian (second-order derivative of objective), summed over observations that fall in this node.
3840
3841
            - ``count`` : int64, number of records in the training data that fall into this node.

3842
3843
3844
3845
3846
3847
        Returns
        -------
        result : pandas DataFrame
            Returns a pandas DataFrame of the parsed model.
        """
        if not PANDAS_INSTALLED:
3848
3849
3850
3851
            raise LightGBMError(
                "This method cannot be run without pandas installed. "
                "You must install pandas and restart your session to use this method."
            )
3852
3853

        if self.num_trees() == 0:
3854
            raise LightGBMError("There are no trees in this Booster and thus nothing to parse")
3855

3856
        def _is_split_node(tree: Dict[str, Any]) -> bool:
3857
            return "split_index" in tree.keys()
3858

3859
3860
3861
3862
3863
        def create_node_record(
            tree: Dict[str, Any],
            node_depth: int = 1,
            tree_index: Optional[int] = None,
            feature_names: Optional[List[str]] = None,
3864
            parent_node: Optional[str] = None,
3865
3866
3867
        ) -> Dict[str, Any]:
            def _get_node_index(
                tree: Dict[str, Any],
3868
                tree_index: Optional[int],
3869
            ) -> str:
3870
                tree_num = f"{tree_index}-" if tree_index is not None else ""
3871
                is_split = _is_split_node(tree)
3872
                node_type = "S" if is_split else "L"
3873
                # if a single node tree it won't have `leaf_index` so return 0
3874
                node_num = tree.get("split_index" if is_split else "leaf_index", 0)
3875
                return f"{tree_num}{node_type}{node_num}"
3876

3877
3878
            def _get_split_feature(
                tree: Dict[str, Any],
3879
                feature_names: Optional[List[str]],
3880
            ) -> Optional[str]:
3881
3882
                if _is_split_node(tree):
                    if feature_names is not None:
3883
                        feature_name = feature_names[tree["split_feature"]]
3884
                    else:
3885
                        feature_name = tree["split_feature"]
3886
3887
3888
3889
                else:
                    feature_name = None
                return feature_name

3890
            def _is_single_node_tree(tree: Dict[str, Any]) -> bool:
3891
                return set(tree.keys()) == {"leaf_value", "leaf_count"}
3892
3893

            # Create the node record, and populate universal data members
3894
            node: Dict[str, Union[int, str, None]] = OrderedDict()
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
            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
3910
3911
3912

            # Update values to reflect node type (leaf or split)
            if _is_split_node(tree):
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
                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"]
3923
            else:
3924
                node["value"] = tree["leaf_value"]
3925
                if not _is_single_node_tree(tree):
3926
3927
                    node["weight"] = tree["leaf_weight"]
                    node["count"] = tree["leaf_count"]
3928
3929
3930

            return node

3931
3932
3933
3934
3935
        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,
3936
            parent_node: Optional[str] = None,
3937
        ) -> List[Dict[str, Any]]:
3938
3939
3940
3941
3942
3943
3944
            node = create_node_record(
                tree=tree,
                node_depth=node_depth,
                tree_index=tree_index,
                feature_names=feature_names,
                parent_node=parent_node,
            )
3945
3946
3947
3948
3949

            res = [node]

            if _is_split_node(tree):
                # traverse the next level of the tree
3950
                children = ["left_child", "right_child"]
3951
3952
                for child in children:
                    subtree_list = tree_dict_to_node_list(
3953
                        tree=tree[child],
3954
3955
3956
                        node_depth=node_depth + 1,
                        tree_index=tree_index,
                        feature_names=feature_names,
3957
                        parent_node=node["node_index"],
3958
                    )
3959
3960
3961
3962
3963
3964
                    # 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()
3965
        feature_names = model_dict["feature_names"]
3966
        model_list = []
3967
3968
3969
3970
3971
3972
        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
                )
            )
3973

3974
        return pd_DataFrame(model_list, columns=model_list[0].keys())
3975

3976
    def set_train_data_name(self, name: str) -> "Booster":
3977
3978
3979
3980
        """Set the name to the training Dataset.

        Parameters
        ----------
3981
        name : str
Nikita Titov's avatar
Nikita Titov committed
3982
3983
3984
3985
3986
3987
            Name for the training Dataset.

        Returns
        -------
        self : Booster
            Booster with set training Dataset name.
3988
        """
3989
        self._train_data_name = name
Nikita Titov's avatar
Nikita Titov committed
3990
        return self
wxchan's avatar
wxchan committed
3991

3992
    def add_valid(self, data: Dataset, name: str) -> "Booster":
3993
        """Add validation data.
wxchan's avatar
wxchan committed
3994
3995
3996
3997

        Parameters
        ----------
        data : Dataset
3998
            Validation data.
3999
        name : str
4000
            Name of validation data.
Nikita Titov's avatar
Nikita Titov committed
4001
4002
4003
4004
4005

        Returns
        -------
        self : Booster
            Booster with set validation data.
wxchan's avatar
wxchan committed
4006
        """
Guolin Ke's avatar
Guolin Ke committed
4007
        if not isinstance(data, Dataset):
4008
            raise TypeError(f"Validation data should be Dataset instance, met {type(data).__name__}")
Guolin Ke's avatar
Guolin Ke committed
4009
        if data._predictor is not self.__init_predictor:
4010
            raise LightGBMError("Add validation data failed, you should use same predictor for these data")
4011
4012
4013
4014
4015
4016
        _safe_call(
            _LIB.LGBM_BoosterAddValidData(
                self._handle,
                data.construct()._handle,
            )
        )
wxchan's avatar
wxchan committed
4017
4018
4019
4020
4021
        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
4022
        return self
wxchan's avatar
wxchan committed
4023

4024
    def reset_parameter(self, params: Dict[str, Any]) -> "Booster":
4025
        """Reset parameters of Booster.
wxchan's avatar
wxchan committed
4026
4027
4028
4029

        Parameters
        ----------
        params : dict
4030
            New parameters for Booster.
Nikita Titov's avatar
Nikita Titov committed
4031
4032
4033
4034
4035

        Returns
        -------
        self : Booster
            Booster with new parameters.
wxchan's avatar
wxchan committed
4036
        """
4037
        params_str = _param_dict_to_str(params)
wxchan's avatar
wxchan committed
4038
        if params_str:
4039
4040
4041
4042
4043
4044
            _safe_call(
                _LIB.LGBM_BoosterResetParameter(
                    self._handle,
                    _c_str(params_str),
                )
            )
Guolin Ke's avatar
Guolin Ke committed
4045
        self.params.update(params)
Nikita Titov's avatar
Nikita Titov committed
4046
        return self
wxchan's avatar
wxchan committed
4047

4048
4049
4050
    def update(
        self,
        train_set: Optional[Dataset] = None,
4051
        fobj: Optional[_LGBM_CustomObjectiveFunction] = None,
4052
    ) -> bool:
Nikita Titov's avatar
Nikita Titov committed
4053
        """Update Booster for one iteration.
4054

wxchan's avatar
wxchan committed
4055
4056
        Parameters
        ----------
4057
4058
4059
4060
        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
4061
            Customized objective function.
4062
4063
4064
            Should accept two parameters: preds, train_data,
            and return (grad, hess).

4065
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4066
                    The predicted values.
4067
4068
                    Predicted values are returned before any transformation,
                    e.g. they are raw margin instead of probability of positive class for binary task.
4069
4070
                train_data : Dataset
                    The training dataset.
4071
                grad : numpy 1-D array or numpy 2-D array (for multi-class task)
4072
4073
                    The value of the first order derivative (gradient) of the loss
                    with respect to the elements of preds for each sample point.
4074
                hess : numpy 1-D array or numpy 2-D array (for multi-class task)
4075
4076
                    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
4077

4078
            For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes],
4079
            and grad and hess should be returned in the same format.
4080

wxchan's avatar
wxchan committed
4081
4082
        Returns
        -------
4083
4084
4085
4086
4087
4088
4089
        produced_empty_tree : bool
            ``True`` if the tree(s) produced by this iteration did not have any splits.
            This usually means that training is "finished" (calling ``update()`` again
            will not change the model's predictions). However, that is not always the
            case. For example, if you have added any randomness (like column sampling by
            setting ``feature_fraction_bynode < 1.0``), it is possible that another call
            to ``update()`` would produce a non-empty tree.
wxchan's avatar
wxchan committed
4090
        """
4091
        # need reset training data
4092
4093
4094
4095
4096
4097
        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
4098
            if not isinstance(train_set, Dataset):
4099
                raise TypeError(f"Training data should be Dataset instance, met {type(train_set).__name__}")
Guolin Ke's avatar
Guolin Ke committed
4100
            if train_set._predictor is not self.__init_predictor:
4101
                raise LightGBMError("Replace training data failed, you should use same predictor for these data")
wxchan's avatar
wxchan committed
4102
            self.train_set = train_set
4103
4104
4105
4106
4107
4108
            _safe_call(
                _LIB.LGBM_BoosterResetTrainingData(
                    self._handle,
                    self.train_set.construct()._handle,
                )
            )
wxchan's avatar
wxchan committed
4109
            self.__inner_predict_buffer[0] = None
4110
            self.train_set_version = self.train_set.version
4111
        produced_empty_tree = ctypes.c_int(0)
wxchan's avatar
wxchan committed
4112
        if fobj is None:
4113
            if self.__set_objective_to_none:
4114
4115
4116
4117
                raise LightGBMError("Cannot update due to null objective function.")
            _safe_call(
                _LIB.LGBM_BoosterUpdateOneIter(
                    self._handle,
4118
                    ctypes.byref(produced_empty_tree),
4119
4120
                )
            )
4121
            self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
4122
            return produced_empty_tree.value == 1
wxchan's avatar
wxchan committed
4123
        else:
4124
            if not self.__set_objective_to_none:
Nikita Titov's avatar
Nikita Titov committed
4125
                self.reset_parameter({"objective": "none"}).__set_objective_to_none = True
wxchan's avatar
wxchan committed
4126
4127
4128
            grad, hess = fobj(self.__inner_predict(0), self.train_set)
            return self.__boost(grad, hess)

4129
4130
4131
    def __boost(
        self,
        grad: np.ndarray,
4132
        hess: np.ndarray,
4133
    ) -> bool:
4134
        """Boost Booster for one iteration with customized gradient statistics.
Nikita Titov's avatar
Nikita Titov committed
4135

Nikita Titov's avatar
Nikita Titov committed
4136
4137
        .. note::

4138
4139
            Score is returned before any transformation,
            e.g. it is raw margin instead of probability of positive class for binary task.
4140
            For multi-class task, score are numpy 2-D array of shape = [n_samples, n_classes],
4141
            and grad and hess should be returned in the same format.
4142

wxchan's avatar
wxchan committed
4143
4144
        Parameters
        ----------
4145
        grad : numpy 1-D array or numpy 2-D array (for multi-class task)
4146
4147
            The value of the first order derivative (gradient) of the loss
            with respect to the elements of score for each sample point.
4148
        hess : numpy 1-D array or numpy 2-D array (for multi-class task)
4149
4150
            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
4151
4152
4153

        Returns
        -------
4154
4155
4156
4157
4158
4159
4160
        produced_empty_tree : bool
            ``True`` if the tree(s) produced by this iteration did not have any splits.
            This usually means that training is "finished" (calling ``__boost()`` again
            will not change the model's predictions). However, that is not always the
            case. For example, if you have added any randomness (like column sampling by
            setting ``feature_fraction_bynode < 1.0``), it is possible that another call
            to ``__boost()`` would produce a non-empty tree.
wxchan's avatar
wxchan committed
4161
        """
4162
        if self.__num_class > 1:
4163
4164
4165
4166
            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")
4167
4168
        assert grad.flags.c_contiguous
        assert hess.flags.c_contiguous
wxchan's avatar
wxchan committed
4169
        if len(grad) != len(hess):
4170
4171
            raise ValueError(f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) don't match")
        num_train_data = self.train_set.num_data()
4172
        if len(grad) != num_train_data * self.__num_class:
4173
4174
4175
            raise ValueError(
                f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) "
                f"don't match training data length ({num_train_data}) * "
4176
                f"number of models per one iteration ({self.__num_class})"
4177
            )
4178
        produced_empty_tree = ctypes.c_int(0)
4179
4180
4181
4182
4183
        _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)),
4184
                ctypes.byref(produced_empty_tree),
4185
4186
            )
        )
4187
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
4188
        return produced_empty_tree.value == 1
wxchan's avatar
wxchan committed
4189

4190
    def rollback_one_iter(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
4191
4192
4193
4194
4195
4196
4197
        """Rollback one iteration.

        Returns
        -------
        self : Booster
            Booster with rolled back one iteration.
        """
4198
        _safe_call(_LIB.LGBM_BoosterRollbackOneIter(self._handle))
4199
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
Nikita Titov's avatar
Nikita Titov committed
4200
        return self
wxchan's avatar
wxchan committed
4201

4202
    def current_iteration(self) -> int:
4203
4204
4205
4206
4207
4208
4209
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
Guolin Ke's avatar
Guolin Ke committed
4210
        out_cur_iter = ctypes.c_int(0)
4211
4212
4213
4214
4215
4216
        _safe_call(
            _LIB.LGBM_BoosterGetCurrentIteration(
                self._handle,
                ctypes.byref(out_cur_iter),
            )
        )
wxchan's avatar
wxchan committed
4217
4218
        return out_cur_iter.value

4219
    def num_model_per_iteration(self) -> int:
4220
4221
4222
4223
4224
4225
4226
4227
        """Get number of models per iteration.

        Returns
        -------
        model_per_iter : int
            The number of models per iteration.
        """
        model_per_iter = ctypes.c_int(0)
4228
4229
4230
4231
4232
4233
        _safe_call(
            _LIB.LGBM_BoosterNumModelPerIteration(
                self._handle,
                ctypes.byref(model_per_iter),
            )
        )
4234
4235
        return model_per_iter.value

4236
    def num_trees(self) -> int:
4237
4238
4239
4240
4241
4242
4243
4244
        """Get number of weak sub-models.

        Returns
        -------
        num_trees : int
            The number of weak sub-models.
        """
        num_trees = ctypes.c_int(0)
4245
4246
4247
4248
4249
4250
        _safe_call(
            _LIB.LGBM_BoosterNumberOfTotalModel(
                self._handle,
                ctypes.byref(num_trees),
            )
        )
4251
4252
        return num_trees.value

4253
    def upper_bound(self) -> float:
4254
4255
4256
4257
        """Get upper bound value of a model.

        Returns
        -------
4258
        upper_bound : float
4259
4260
4261
            Upper bound value of the model.
        """
        ret = ctypes.c_double(0)
4262
4263
4264
4265
4266
4267
        _safe_call(
            _LIB.LGBM_BoosterGetUpperBoundValue(
                self._handle,
                ctypes.byref(ret),
            )
        )
4268
4269
        return ret.value

4270
    def lower_bound(self) -> float:
4271
4272
4273
4274
        """Get lower bound value of a model.

        Returns
        -------
4275
        lower_bound : float
4276
4277
4278
            Lower bound value of the model.
        """
        ret = ctypes.c_double(0)
4279
4280
4281
4282
4283
4284
        _safe_call(
            _LIB.LGBM_BoosterGetLowerBoundValue(
                self._handle,
                ctypes.byref(ret),
            )
        )
4285
4286
        return ret.value

4287
4288
4289
4290
    def eval(
        self,
        data: Dataset,
        name: str,
4291
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4292
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4293
        """Evaluate for data.
wxchan's avatar
wxchan committed
4294
4295
4296

        Parameters
        ----------
4297
4298
        data : Dataset
            Data for the evaluating.
4299
        name : str
4300
            Name of the data.
4301
        feval : callable, list of callable, or None, optional (default=None)
4302
            Customized evaluation function.
4303
            Each evaluation function should accept two parameters: preds, eval_data,
4304
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4305

4306
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4307
                    The predicted values.
4308
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4309
                    If custom objective function is used, predicted values are returned before any transformation,
4310
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
4311
                eval_data : Dataset
4312
                    A ``Dataset`` to evaluate.
4313
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4314
                    The name of evaluation function (without whitespace).
4315
4316
4317
4318
4319
                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
4320
4321
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4322
        result : list
4323
            List with (dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4324
        """
Guolin Ke's avatar
Guolin Ke committed
4325
4326
        if not isinstance(data, Dataset):
            raise TypeError("Can only eval for Dataset instance")
wxchan's avatar
wxchan committed
4327
4328
4329
4330
        data_idx = -1
        if data is self.train_set:
            data_idx = 0
        else:
4331
            for i in range(len(self.valid_sets)):
wxchan's avatar
wxchan committed
4332
4333
4334
                if data is self.valid_sets[i]:
                    data_idx = i + 1
                    break
4335
        # need to push new valid data
wxchan's avatar
wxchan committed
4336
4337
4338
4339
        if data_idx == -1:
            self.add_valid(data, name)
            data_idx = self.__num_dataset - 1

4340
        return self.__inner_eval(data_name=name, data_idx=data_idx, feval=feval)
wxchan's avatar
wxchan committed
4341

4342
4343
    def eval_train(
        self,
4344
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4345
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4346
        """Evaluate for training data.
wxchan's avatar
wxchan committed
4347
4348
4349

        Parameters
        ----------
4350
        feval : callable, list of callable, or None, optional (default=None)
4351
            Customized evaluation function.
4352
            Each evaluation function should accept two parameters: preds, eval_data,
4353
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4354

4355
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4356
                    The predicted values.
4357
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4358
                    If custom objective function is used, predicted values are returned before any transformation,
4359
                    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
4360
                eval_data : Dataset
4361
                    The training dataset.
4362
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4363
                    The name of evaluation function (without whitespace).
4364
4365
4366
4367
4368
                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
4369
4370
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4371
        result : list
4372
            List with (train_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4373
        """
4374
        return self.__inner_eval(data_name=self._train_data_name, data_idx=0, feval=feval)
wxchan's avatar
wxchan committed
4375

4376
4377
    def eval_valid(
        self,
4378
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None,
4379
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4380
        """Evaluate for validation data.
wxchan's avatar
wxchan committed
4381
4382
4383

        Parameters
        ----------
4384
        feval : callable, list of callable, or None, optional (default=None)
4385
            Customized evaluation function.
4386
            Each evaluation function should accept two parameters: preds, eval_data,
4387
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4388

4389
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4390
                    The predicted values.
4391
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4392
                    If custom objective function is used, predicted values are returned before any transformation,
4393
                    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
4394
                eval_data : Dataset
4395
                    The validation dataset.
4396
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4397
                    The name of evaluation function (without whitespace).
4398
4399
4400
4401
4402
                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
4403
4404
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4405
        result : list
4406
            List with (validation_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4407
        """
4408
4409
4410
        return [
            item
            for i in range(1, self.__num_dataset)
4411
            for item in self.__inner_eval(data_name=self.name_valid_sets[i - 1], data_idx=i, feval=feval)
4412
        ]
wxchan's avatar
wxchan committed
4413

4414
4415
4416
4417
4418
    def save_model(
        self,
        filename: Union[str, Path],
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4419
        importance_type: str = "split",
4420
    ) -> "Booster":
4421
        """Save Booster to file.
wxchan's avatar
wxchan committed
4422
4423
4424

        Parameters
        ----------
4425
        filename : str or pathlib.Path
4426
            Filename to save Booster.
4427
4428
4429
4430
        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
4431
        start_iteration : int, optional (default=0)
4432
            Start index of the iteration that should be saved.
4433
        importance_type : str, optional (default="split")
4434
4435
4436
            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
4437
4438
4439
4440
4441

        Returns
        -------
        self : Booster
            Returns self.
wxchan's avatar
wxchan committed
4442
        """
4443
        if num_iteration is None:
4444
            num_iteration = self.best_iteration
4445
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4446
4447
4448
4449
4450
4451
4452
4453
4454
        _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)),
            )
        )
4455
        _dump_pandas_categorical(self.pandas_categorical, filename)
Nikita Titov's avatar
Nikita Titov committed
4456
        return self
wxchan's avatar
wxchan committed
4457

4458
4459
4460
    def shuffle_models(
        self,
        start_iteration: int = 0,
4461
        end_iteration: int = -1,
4462
    ) -> "Booster":
4463
        """Shuffle models.
Nikita Titov's avatar
Nikita Titov committed
4464

4465
4466
4467
        Parameters
        ----------
        start_iteration : int, optional (default=0)
4468
            The first iteration that will be shuffled.
4469
4470
        end_iteration : int, optional (default=-1)
            The last iteration that will be shuffled.
4471
            If <= 0, means the last available iteration.
4472

Nikita Titov's avatar
Nikita Titov committed
4473
4474
4475
4476
        Returns
        -------
        self : Booster
            Booster with shuffled models.
4477
        """
4478
4479
4480
4481
4482
4483
4484
        _safe_call(
            _LIB.LGBM_BoosterShuffleModels(
                self._handle,
                ctypes.c_int(start_iteration),
                ctypes.c_int(end_iteration),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
4485
        return self
4486

4487
    def model_from_string(self, model_str: str) -> "Booster":
4488
4489
4490
4491
        """Load Booster from a string.

        Parameters
        ----------
4492
        model_str : str
4493
4494
4495
4496
            Model will be loaded from this string.

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4497
        self : Booster
4498
4499
            Loaded Booster object.
        """
4500
4501
4502
        # ensure that existing Booster is freed before replacing it
        # with a new one createdfrom file
        _safe_call(_LIB.LGBM_BoosterFree(self._handle))
4503
        self._free_buffer()
4504
        self._handle = ctypes.c_void_p()
4505
        out_num_iterations = ctypes.c_int(0)
4506
4507
4508
4509
4510
4511
4512
        _safe_call(
            _LIB.LGBM_BoosterLoadModelFromString(
                _c_str(model_str),
                ctypes.byref(out_num_iterations),
                ctypes.byref(self._handle),
            )
        )
4513
        out_num_class = ctypes.c_int(0)
4514
4515
4516
4517
4518
4519
        _safe_call(
            _LIB.LGBM_BoosterGetNumClasses(
                self._handle,
                ctypes.byref(out_num_class),
            )
        )
4520
        self.__num_class = out_num_class.value
4521
        self.pandas_categorical = _load_pandas_categorical(model_str=model_str)
4522
4523
        return self

4524
4525
4526
4527
    def model_to_string(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4528
        importance_type: str = "split",
4529
    ) -> str:
4530
        """Save Booster to string.
4531

4532
4533
4534
4535
4536
4537
        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
4538
        start_iteration : int, optional (default=0)
4539
            Start index of the iteration that should be saved.
4540
        importance_type : str, optional (default="split")
4541
4542
4543
            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.
4544
4545
4546

        Returns
        -------
4547
        str_repr : str
4548
4549
            String representation of Booster.
        """
4550
        if num_iteration is None:
4551
            num_iteration = self.best_iteration
4552
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4553
        buffer_len = 1 << 20
4554
        tmp_out_len = ctypes.c_int64(0)
4555
        string_buffer = ctypes.create_string_buffer(buffer_len)
4556
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4557
4558
        _safe_call(
            _LIB.LGBM_BoosterSaveModelToString(
4559
                self._handle,
4560
                ctypes.c_int(start_iteration),
4561
                ctypes.c_int(num_iteration),
4562
                ctypes.c_int(importance_type_int),
4563
                ctypes.c_int64(buffer_len),
4564
                ctypes.byref(tmp_out_len),
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
                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")
4585
4586
        ret += _dump_pandas_categorical(self.pandas_categorical)
        return ret
4587

4588
4589
4590
4591
    def dump_model(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
4592
4593
        importance_type: str = "split",
        object_hook: Optional[Callable[[Dict[str, Any]], Dict[str, Any]]] = None,
4594
    ) -> Dict[str, Any]:
Nikita Titov's avatar
Nikita Titov committed
4595
        """Dump Booster to JSON format.
wxchan's avatar
wxchan committed
4596

4597
4598
        Parameters
        ----------
4599
4600
4601
4602
        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
4603
        start_iteration : int, optional (default=0)
4604
            Start index of the iteration that should be dumped.
4605
        importance_type : str, optional (default="split")
4606
4607
4608
            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.
4609
4610
4611
4612
4613
4614
4615
4616
4617
        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.
4618

wxchan's avatar
wxchan committed
4619
4620
        Returns
        -------
4621
        json_repr : dict
Nikita Titov's avatar
Nikita Titov committed
4622
            JSON format of Booster.
wxchan's avatar
wxchan committed
4623
        """
4624
        if num_iteration is None:
4625
            num_iteration = self.best_iteration
4626
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
4627
        buffer_len = 1 << 20
4628
        tmp_out_len = ctypes.c_int64(0)
wxchan's avatar
wxchan committed
4629
        string_buffer = ctypes.create_string_buffer(buffer_len)
4630
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4631
4632
        _safe_call(
            _LIB.LGBM_BoosterDumpModel(
4633
                self._handle,
4634
                ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
4635
                ctypes.c_int(num_iteration),
4636
                ctypes.c_int(importance_type_int),
4637
                ctypes.c_int64(buffer_len),
wxchan's avatar
wxchan committed
4638
                ctypes.byref(tmp_out_len),
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
                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,
            )
        )
4665
        return ret
wxchan's avatar
wxchan committed
4666

4667
4668
    def predict(
        self,
4669
        data: _LGBM_PredictDataType,
4670
4671
4672
4673
4674
4675
4676
        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,
4677
        **kwargs: Any,
4678
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
4679
        """Make a prediction.
wxchan's avatar
wxchan committed
4680
4681
4682

        Parameters
        ----------
4683
        data : str, pathlib.Path, numpy array, pandas DataFrame, scipy.sparse or pyarrow Table
4684
            Data source for prediction.
4685
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4686
        start_iteration : int, optional (default=0)
4687
            Start index of the iteration to predict.
4688
            If <= 0, starts from the first iteration.
4689
        num_iteration : int or None, optional (default=None)
4690
4691
4692
4693
            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).
4694
4695
4696
4697
        raw_score : bool, optional (default=False)
            Whether to predict raw scores.
        pred_leaf : bool, optional (default=False)
            Whether to predict leaf index.
4698
4699
        pred_contrib : bool, optional (default=False)
            Whether to predict feature contributions.
4700

Nikita Titov's avatar
Nikita Titov committed
4701
4702
4703
4704
4705
4706
4707
            .. 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.
4708

4709
4710
        data_has_header : bool, optional (default=False)
            Whether the data has header.
4711
            Used only if data is str.
4712
4713
4714
        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.
4715
4716
        **kwargs
            Other parameters for the prediction.
wxchan's avatar
wxchan committed
4717
4718
4719

        Returns
        -------
4720
        result : numpy array, scipy.sparse or list of scipy.sparse
4721
            Prediction result.
4722
            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
4723
        """
4724
4725
4726
4727
        predictor = _InnerPredictor.from_booster(
            booster=self,
            pred_parameter=deepcopy(kwargs),
        )
4728
        if num_iteration is None:
4729
            if start_iteration <= 0:
4730
4731
4732
                num_iteration = self.best_iteration
            else:
                num_iteration = -1
4733
4734
4735
4736
4737
4738
4739
4740
        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,
4741
            validate_features=validate_features,
4742
        )
wxchan's avatar
wxchan committed
4743

4744
4745
    def refit(
        self,
4746
        data: _LGBM_TrainDataType,
4747
        label: _LGBM_LabelType,
4748
4749
        decay_rate: float = 0.9,
        reference: Optional[Dataset] = None,
4750
4751
4752
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
4753
4754
        feature_name: _LGBM_FeatureNameConfiguration = "auto",
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = "auto",
4755
4756
4757
        dataset_params: Optional[Dict[str, Any]] = None,
        free_raw_data: bool = True,
        validate_features: bool = False,
4758
        **kwargs: Any,
4759
    ) -> "Booster":
Guolin Ke's avatar
Guolin Ke committed
4760
4761
4762
4763
        """Refit the existing Booster by new data.

        Parameters
        ----------
4764
        data : str, pathlib.Path, numpy array, pandas DataFrame, scipy.sparse, Sequence, list of Sequence, list of numpy array or pyarrow Table
Guolin Ke's avatar
Guolin Ke committed
4765
            Data source for refit.
4766
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4767
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array or pyarrow ChunkedArray
Guolin Ke's avatar
Guolin Ke committed
4768
4769
            Label for refit.
        decay_rate : float, optional (default=0.9)
4770
4771
            Decay rate of refit,
            will use ``leaf_output = decay_rate * old_leaf_output + (1.0 - decay_rate) * new_leaf_output`` to refit trees.
4772
4773
        reference : Dataset or None, optional (default=None)
            Reference for ``data``.
4774
4775
4776

            .. versionadded:: 4.0.0

4777
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4778
            Weight for each ``data`` instance. Weights should be non-negative.
4779
4780
4781

            .. versionadded:: 4.0.0

4782
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4783
4784
4785
4786
4787
            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.
4788
4789
4790

            .. versionadded:: 4.0.0

4791
        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)
4792
            Init score for ``data``.
4793
4794
4795

            .. versionadded:: 4.0.0

4796
4797
4798
        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.
4799
4800
4801

            .. versionadded:: 4.0.0

4802
4803
4804
4805
4806
        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.
4807
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
4808
4809
4810
            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.
4811
            Floating point numbers in categorical features will be rounded towards 0.
4812
4813
4814

            .. versionadded:: 4.0.0

4815
4816
        dataset_params : dict or None, optional (default=None)
            Other parameters for Dataset ``data``.
4817
4818
4819

            .. versionadded:: 4.0.0

4820
4821
        free_raw_data : bool, optional (default=True)
            If True, raw data is freed after constructing inner Dataset for ``data``.
4822
4823
4824

            .. versionadded:: 4.0.0

4825
4826
4827
        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.
4828
4829
4830

            .. versionadded:: 4.0.0

4831
4832
        **kwargs
            Other parameters for refit.
4833
            These parameters will be passed to ``predict`` method.
Guolin Ke's avatar
Guolin Ke committed
4834
4835
4836
4837
4838
4839

        Returns
        -------
        result : Booster
            Refitted Booster.
        """
4840
        if self.__set_objective_to_none:
4841
            raise LightGBMError("Cannot refit due to null objective function.")
4842
4843
        if dataset_params is None:
            dataset_params = {}
4844
        predictor = _InnerPredictor.from_booster(booster=self, pred_parameter=deepcopy(kwargs))
4845
        leaf_preds: np.ndarray = predictor.predict(  # type: ignore[assignment]
4846
4847
4848
            data=data,
            start_iteration=-1,
            pred_leaf=True,
4849
            validate_features=validate_features,
4850
        )
4851
        nrow, ncol = leaf_preds.shape
4852
        out_is_linear = ctypes.c_int(0)
4853
4854
4855
4856
4857
4858
        _safe_call(
            _LIB.LGBM_BoosterGetLinear(
                self._handle,
                ctypes.byref(out_is_linear),
            )
        )
Nikita Titov's avatar
Nikita Titov committed
4859
4860
4861
        new_params = _choose_param_value(
            main_param_name="linear_tree",
            params=self.params,
4862
            default_value=None,
Nikita Titov's avatar
Nikita Titov committed
4863
        )
4864
        new_params["linear_tree"] = bool(out_is_linear.value)
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
        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,
        )
4878
        new_params["refit_decay_rate"] = decay_rate
4879
        new_booster = Booster(new_params, train_set)
Guolin Ke's avatar
Guolin Ke committed
4880
        # Copy models
4881
4882
4883
4884
4885
4886
        _safe_call(
            _LIB.LGBM_BoosterMerge(
                new_booster._handle,
                predictor._handle,
            )
        )
Guolin Ke's avatar
Guolin Ke committed
4887
        leaf_preds = leaf_preds.reshape(-1)
4888
        ptr_data, _, _ = _c_int_array(leaf_preds)
4889
4890
4891
4892
4893
4894
4895
4896
        _safe_call(
            _LIB.LGBM_BoosterRefit(
                new_booster._handle,
                ptr_data,
                ctypes.c_int32(nrow),
                ctypes.c_int32(ncol),
            )
        )
4897
        new_booster._network = self._network
Guolin Ke's avatar
Guolin Ke committed
4898
4899
        return new_booster

4900
    def get_leaf_output(self, tree_id: int, leaf_id: int) -> float:
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
        """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.
        """
4915
        ret = ctypes.c_double(0)
4916
4917
4918
4919
4920
4921
4922
4923
        _safe_call(
            _LIB.LGBM_BoosterGetLeafValue(
                self._handle,
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
                ctypes.byref(ret),
            )
        )
4924
4925
        return ret.value

4926
4927
4928
4929
4930
    def set_leaf_output(
        self,
        tree_id: int,
        leaf_id: int,
        value: float,
4931
    ) -> "Booster":
4932
4933
        """Set the output of a leaf.

4934
4935
        .. versionadded:: 4.0.0

4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
        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(
4952
                self._handle,
4953
4954
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
4955
                ctypes.c_double(value),
4956
4957
4958
4959
            )
        )
        return self

4960
    def num_feature(self) -> int:
4961
4962
4963
4964
4965
4966
4967
        """Get number of features.

        Returns
        -------
        num_feature : int
            The number of features.
        """
4968
        out_num_feature = ctypes.c_int(0)
4969
4970
4971
4972
4973
4974
        _safe_call(
            _LIB.LGBM_BoosterGetNumFeature(
                self._handle,
                ctypes.byref(out_num_feature),
            )
        )
4975
4976
        return out_num_feature.value

4977
    def feature_name(self) -> List[str]:
4978
        """Get names of features.
wxchan's avatar
wxchan committed
4979
4980
4981

        Returns
        -------
4982
        result : list of str
4983
            List with names of features.
wxchan's avatar
wxchan committed
4984
        """
4985
        num_feature = self.num_feature()
4986
        # Get name of features
wxchan's avatar
wxchan committed
4987
        tmp_out_len = ctypes.c_int(0)
4988
4989
        reserved_string_buffer_size = 255
        required_string_buffer_size = ctypes.c_size_t(0)
4990
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
4991
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
        _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
5002
5003
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
5004
5005
5006
5007
        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)]
5008
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
            _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
5020

5021
5022
    def feature_importance(
        self,
5023
5024
        importance_type: str = "split",
        iteration: Optional[int] = None,
5025
    ) -> np.ndarray:
5026
        """Get feature importances.
5027

5028
5029
        Parameters
        ----------
5030
        importance_type : str, optional (default="split")
5031
5032
5033
            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.
5034
5035
5036
5037
        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).
5038

5039
5040
        Returns
        -------
5041
5042
        result : numpy array
            Array with feature importances.
5043
        """
5044
5045
        if iteration is None:
            iteration = self.best_iteration
5046
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
5047
        result = np.empty(self.num_feature(), dtype=np.float64)
5048
5049
5050
5051
5052
5053
5054
5055
        _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)),
            )
        )
5056
        if importance_type_int == _C_API_FEATURE_IMPORTANCE_SPLIT:
5057
            return result.astype(np.int32)
5058
5059
        else:
            return result
5060

5061
5062
5063
5064
    def get_split_value_histogram(
        self,
        feature: Union[int, str],
        bins: Optional[Union[int, str]] = None,
5065
        xgboost_style: bool = False,
5066
    ) -> Union[Tuple[np.ndarray, np.ndarray], np.ndarray, pd_DataFrame]:
5067
5068
5069
5070
        """Get split value histogram for the specified feature.

        Parameters
        ----------
5071
        feature : int or str
5072
5073
            The feature name or index the histogram is calculated for.
            If int, interpreted as index.
5074
            If str, interpreted as name.
5075

Nikita Titov's avatar
Nikita Titov committed
5076
5077
5078
            .. warning::

                Categorical features are not supported.
5079

5080
        bins : int, str or None, optional (default=None)
5081
5082
5083
            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.
5084
            If str, it should be one from the list of the supported values by ``numpy.histogram()`` function.
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
        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.
        """
5099

5100
        def add(root: Dict[str, Any]) -> None:
5101
            """Recursively add thresholds."""
5102
            if "split_index" in root:  # non-leaf
5103
                if feature_names is not None and isinstance(feature, str):
5104
                    split_feature = feature_names[root["split_feature"]]
5105
                else:
5106
                    split_feature = root["split_feature"]
5107
                if split_feature == feature:
5108
5109
                    if isinstance(root["threshold"], str):
                        raise LightGBMError("Cannot compute split value histogram for the categorical feature")
5110
                    values.append(root["threshold"])
5111
5112
                add(root["left_child"])
                add(root["right_child"])
5113
5114

        model = self.dump_model()
5115
5116
        feature_names = model.get("feature_names")
        tree_infos = model["tree_info"]
5117
        values: List[float] = []
5118
        for tree_info in tree_infos:
5119
            add(tree_info["tree_structure"])
5120

5121
        if bins is None or isinstance(bins, int) and xgboost_style:
5122
5123
5124
5125
5126
5127
5128
            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:
5129
                return pd_DataFrame(ret, columns=["SplitValue", "Count"])
5130
5131
5132
5133
5134
            else:
                return ret
        else:
            return hist, bin_edges

5135
5136
    def __inner_eval(
        self,
5137
        *,
5138
5139
        data_name: str,
        data_idx: int,
5140
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]],
5141
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
5142
        """Evaluate training or validation data."""
wxchan's avatar
wxchan committed
5143
        if data_idx >= self.__num_dataset:
5144
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
5145
5146
5147
        self.__get_eval_info()
        ret = []
        if self.__num_inner_eval > 0:
5148
            result = np.empty(self.__num_inner_eval, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
5149
            tmp_out_len = ctypes.c_int(0)
5150
5151
5152
5153
5154
5155
5156
5157
            _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
5158
            if tmp_out_len.value != self.__num_inner_eval:
5159
                raise ValueError("Wrong length of eval results")
5160
            for i in range(self.__num_inner_eval):
5161
                ret.append((data_name, self.__name_inner_eval[i], result[i], self.__higher_better_inner_eval[i]))
5162
5163
        if callable(feval):
            feval = [feval]
wxchan's avatar
wxchan committed
5164
5165
5166
5167
5168
        if feval is not None:
            if data_idx == 0:
                cur_data = self.train_set
            else:
                cur_data = self.valid_sets[data_idx - 1]
5169
5170
5171
5172
5173
5174
5175
5176
5177
            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
5178
5179
5180
                    ret.append((data_name, eval_name, val, is_higher_better))
        return ret

5181
    def __inner_predict(self, data_idx: int) -> np.ndarray:
5182
        """Predict for training and validation dataset."""
wxchan's avatar
wxchan committed
5183
        if data_idx >= self.__num_dataset:
5184
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
5185
5186
5187
5188
5189
        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
5190
            self.__inner_predict_buffer[data_idx] = np.empty(n_preds, dtype=np.float64)
5191
        # avoid to predict many time in one iteration
wxchan's avatar
wxchan committed
5192
5193
        if not self.__is_predicted_cur_iter[data_idx]:
            tmp_out_len = ctypes.c_int64(0)
5194
            data_ptr = self.__inner_predict_buffer[data_idx].ctypes.data_as(ctypes.POINTER(ctypes.c_double))  # type: ignore[union-attr]
5195
5196
5197
5198
5199
5200
5201
5202
            _safe_call(
                _LIB.LGBM_BoosterGetPredict(
                    self._handle,
                    ctypes.c_int(data_idx),
                    ctypes.byref(tmp_out_len),
                    data_ptr,
                )
            )
5203
            if tmp_out_len.value != len(self.__inner_predict_buffer[data_idx]):  # type: ignore[arg-type]
5204
                raise ValueError(f"Wrong length of predict results for data {data_idx}")
wxchan's avatar
wxchan committed
5205
            self.__is_predicted_cur_iter[data_idx] = True
5206
        result: np.ndarray = self.__inner_predict_buffer[data_idx]  # type: ignore[assignment]
5207
5208
        if self.__num_class > 1:
            num_data = result.size // self.__num_class
5209
            result = result.reshape(num_data, self.__num_class, order="F")
5210
        return result
wxchan's avatar
wxchan committed
5211

5212
    def __get_eval_info(self) -> None:
5213
        """Get inner evaluation count and names."""
wxchan's avatar
wxchan committed
5214
5215
        if self.__need_reload_eval_info:
            self.__need_reload_eval_info = False
Guolin Ke's avatar
Guolin Ke committed
5216
            out_num_eval = ctypes.c_int(0)
5217
            # Get num of inner evals
5218
5219
5220
5221
5222
5223
            _safe_call(
                _LIB.LGBM_BoosterGetEvalCounts(
                    self._handle,
                    ctypes.byref(out_num_eval),
                )
            )
wxchan's avatar
wxchan committed
5224
5225
            self.__num_inner_eval = out_num_eval.value
            if self.__num_inner_eval > 0:
5226
                # Get name of eval metrics
Guolin Ke's avatar
Guolin Ke committed
5227
                tmp_out_len = ctypes.c_int(0)
5228
5229
5230
                reserved_string_buffer_size = 255
                required_string_buffer_size = ctypes.c_size_t(0)
                string_buffers = [
5231
                    ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(self.__num_inner_eval)
5232
                ]
5233
                ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
                _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
5244
                if self.__num_inner_eval != tmp_out_len.value:
5245
                    raise ValueError("Length of eval names doesn't equal with num_evals")
5246
5247
5248
5249
5250
5251
                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)
                    ]
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
                    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)]
5266
                self.__higher_better_inner_eval = [
5267
                    name.startswith(("auc", "ndcg@", "map@", "average_precision")) for name in self.__name_inner_eval
5268
                ]