basic.py 194 KB
Newer Older
wxchan's avatar
wxchan committed
1
# coding: utf-8
2
"""Wrapper for C API of LightGBM."""
3
import abc
wxchan's avatar
wxchan committed
4
import ctypes
5
import inspect
6
import json
wxchan's avatar
wxchan committed
7
import warnings
8
from collections import OrderedDict
9
from copy import deepcopy
10
from enum import Enum
11
from functools import wraps
12
from os import SEEK_END, environ
13
14
from os.path import getsize
from pathlib import Path
15
from tempfile import NamedTemporaryFile
16
from typing import TYPE_CHECKING, Any, Callable, Dict, Iterable, List, Optional, Set, Tuple, Union
wxchan's avatar
wxchan committed
17
18
19
20

import numpy as np
import scipy.sparse

21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
from .compat import (
    PANDAS_INSTALLED,
    PYARROW_INSTALLED,
    arrow_cffi,
    arrow_is_floating,
    arrow_is_integer,
    concat,
    dt_DataTable,
    pa_Array,
    pa_chunked_array,
    pa_ChunkedArray,
    pa_compute,
    pa_Table,
    pd_CategoricalDtype,
    pd_DataFrame,
    pd_Series,
)
wxchan's avatar
wxchan committed
38
39
from .libpath import find_lib_path

40
41
42
if TYPE_CHECKING:
    from typing import Literal

43
44
45
46
47
48
49
    # typing.TypeGuard was only introduced in Python 3.10
    try:
        from typing import TypeGuard
    except ImportError:
        from typing_extensions import TypeGuard


50
51
52
53
54
55
56
57
58
__all__ = [
    'Booster',
    'Dataset',
    'LGBMDeprecationWarning',
    'LightGBMError',
    'register_logger',
    'Sequence',
]

59
_BoosterHandle = ctypes.c_void_p
60
_DatasetHandle = ctypes.c_void_p
61
62
63
64
_ctypes_int_ptr = Union[
    "ctypes._Pointer[ctypes.c_int32]",
    "ctypes._Pointer[ctypes.c_int64]"
]
65
66
67
68
_ctypes_int_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_int32]]",
    "ctypes.Array[ctypes._Pointer[ctypes.c_int64]]"
]
69
70
71
72
73
74
75
76
_ctypes_float_ptr = Union[
    "ctypes._Pointer[ctypes.c_float]",
    "ctypes._Pointer[ctypes.c_double]"
]
_ctypes_float_array = Union[
    "ctypes.Array[ctypes._Pointer[ctypes.c_float]]",
    "ctypes.Array[ctypes._Pointer[ctypes.c_double]]"
]
77
_LGBM_EvalFunctionResultType = Tuple[str, float, bool]
78
_LGBM_BoosterBestScoreType = Dict[str, Dict[str, float]]
79
_LGBM_BoosterEvalMethodResultType = Tuple[str, str, float, bool]
80
_LGBM_BoosterEvalMethodResultWithStandardDeviationType = Tuple[str, str, float, bool, float]
81
82
_LGBM_CategoricalFeatureConfiguration = Union[List[str], List[int], "Literal['auto']"]
_LGBM_FeatureNameConfiguration = Union[List[str], "Literal['auto']"]
83
84
85
86
_LGBM_GroupType = Union[
    List[float],
    List[int],
    np.ndarray,
87
88
89
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
90
]
91
92
93
94
_LGBM_PositionType = Union[
    np.ndarray,
    pd_Series
]
95
96
97
98
99
100
_LGBM_InitScoreType = Union[
    List[float],
    List[List[float]],
    np.ndarray,
    pd_Series,
    pd_DataFrame,
101
102
103
    pa_Table,
    pa_Array,
    pa_ChunkedArray,
104
]
105
106
107
108
109
110
111
112
113
_LGBM_TrainDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
    scipy.sparse.spmatrix,
    "Sequence",
    List["Sequence"],
114
115
    List[np.ndarray],
    pa_Table
116
]
117
_LGBM_LabelType = Union[
118
119
    List[float],
    List[int],
120
121
    np.ndarray,
    pd_Series,
122
123
124
    pd_DataFrame,
    pa_Array,
    pa_ChunkedArray,
125
]
126
127
128
129
130
131
_LGBM_PredictDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
132
133
    scipy.sparse.spmatrix,
    pa_Table,
134
]
135
136
137
138
_LGBM_WeightType = Union[
    List[float],
    List[int],
    np.ndarray,
139
140
141
    pd_Series,
    pa_Array,
    pa_ChunkedArray,
142
]
143
144
145
ZERO_THRESHOLD = 1e-35


146
147
148
149
def _is_zero(x: float) -> bool:
    return -ZERO_THRESHOLD <= x <= ZERO_THRESHOLD


150
def _get_sample_count(total_nrow: int, params: str) -> int:
151
152
153
    sample_cnt = ctypes.c_int(0)
    _safe_call(_LIB.LGBM_GetSampleCount(
        ctypes.c_int32(total_nrow),
154
        _c_str(params),
155
156
157
158
        ctypes.byref(sample_cnt),
    ))
    return sample_cnt.value

wxchan's avatar
wxchan committed
159

160
161
162
163
164
165
class _MissingType(Enum):
    NONE = 'None'
    NAN = 'NaN'
    ZERO = 'Zero'


166
class _DummyLogger:
167
    def info(self, msg: str) -> None:
168
        print(msg)  # noqa: T201
169

170
    def warning(self, msg: str) -> None:
171
172
173
        warnings.warn(msg, stacklevel=3)


174
175
176
_LOGGER: Any = _DummyLogger()
_INFO_METHOD_NAME = "info"
_WARNING_METHOD_NAME = "warning"
177
178


179
180
181
182
def _has_method(logger: Any, method_name: str) -> bool:
    return callable(getattr(logger, method_name, None))


183
184
185
def register_logger(
    logger: Any, info_method_name: str = "info", warning_method_name: str = "warning"
) -> None:
186
187
188
189
    """Register custom logger.

    Parameters
    ----------
190
    logger : Any
191
        Custom logger.
192
193
194
195
    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.
196
    """
197
198
199
200
201
202
    if not _has_method(logger, info_method_name) or not _has_method(logger, warning_method_name):
        raise TypeError(
            f"Logger must provide '{info_method_name}' and '{warning_method_name}' method"
        )

    global _LOGGER, _INFO_METHOD_NAME, _WARNING_METHOD_NAME
203
    _LOGGER = logger
204
205
    _INFO_METHOD_NAME = info_method_name
    _WARNING_METHOD_NAME = warning_method_name
206
207


208
def _normalize_native_string(func: Callable[[str], None]) -> Callable[[str], None]:
209
    """Join log messages from native library which come by chunks."""
210
    msg_normalized: List[str] = []
211
212

    @wraps(func)
213
    def wrapper(msg: str) -> None:
214
215
216
217
218
219
220
221
222
223
224
        nonlocal msg_normalized
        if msg.strip() == '':
            msg = ''.join(msg_normalized)
            msg_normalized = []
            return func(msg)
        else:
            msg_normalized.append(msg)

    return wrapper


225
def _log_info(msg: str) -> None:
226
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
227
228


229
def _log_warning(msg: str) -> None:
230
    getattr(_LOGGER, _WARNING_METHOD_NAME)(msg)
231
232
233


@_normalize_native_string
234
def _log_native(msg: str) -> None:
235
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
236
237


238
def _log_callback(msg: bytes) -> None:
239
    """Redirect logs from native library into Python."""
240
    _log_native(str(msg.decode('utf-8')))
241
242


243
def _load_lib() -> ctypes.CDLL:
244
    """Load LightGBM library."""
wxchan's avatar
wxchan committed
245
246
247
    lib_path = find_lib_path()
    lib = ctypes.cdll.LoadLibrary(lib_path[0])
    lib.LGBM_GetLastError.restype = ctypes.c_char_p
248
    callback = ctypes.CFUNCTYPE(None, ctypes.c_char_p)
249
    lib.callback = callback(_log_callback)  # type: ignore[attr-defined]
250
    if lib.LGBM_RegisterLogCallback(lib.callback) != 0:
251
        raise LightGBMError(lib.LGBM_GetLastError().decode('utf-8'))
wxchan's avatar
wxchan committed
252
253
    return lib

wxchan's avatar
wxchan committed
254

255
256
257
258
259
260
261
# we don't need lib_lightgbm while building docs
_LIB: ctypes.CDLL
if environ.get('LIGHTGBM_BUILD_DOC', False):
    from unittest.mock import Mock  # isort: skip
    _LIB = Mock(ctypes.CDLL)  # type: ignore
else:
    _LIB = _load_lib()
wxchan's avatar
wxchan committed
262

wxchan's avatar
wxchan committed
263

264
_NUMERIC_TYPES = (int, float, bool)
265
266


267
def _safe_call(ret: int) -> None:
268
269
    """Check the return value from C API call.

wxchan's avatar
wxchan committed
270
271
272
    Parameters
    ----------
    ret : int
273
        The return value from C API calls.
wxchan's avatar
wxchan committed
274
275
    """
    if ret != 0:
276
        raise LightGBMError(_LIB.LGBM_GetLastError().decode('utf-8'))
wxchan's avatar
wxchan committed
277

wxchan's avatar
wxchan committed
278

279
def _is_numeric(obj: Any) -> bool:
280
    """Check whether object is a number or not, include numpy number, etc."""
wxchan's avatar
wxchan committed
281
282
283
    try:
        float(obj)
        return True
wxchan's avatar
wxchan committed
284
285
286
    except (TypeError, ValueError):
        # TypeError: obj is not a string or a number
        # ValueError: invalid literal
wxchan's avatar
wxchan committed
287
288
        return False

wxchan's avatar
wxchan committed
289

290
def _is_numpy_1d_array(data: Any) -> bool:
291
    """Check whether data is a numpy 1-D array."""
292
    return isinstance(data, np.ndarray) and len(data.shape) == 1
wxchan's avatar
wxchan committed
293

wxchan's avatar
wxchan committed
294

295
def _is_numpy_column_array(data: Any) -> bool:
296
297
298
299
300
301
302
    """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


303
def _cast_numpy_array_to_dtype(array: np.ndarray, dtype: "np.typing.DTypeLike") -> np.ndarray:
304
    """Cast numpy array to given dtype."""
305
306
307
308
309
    if array.dtype == dtype:
        return array
    return array.astype(dtype=dtype, copy=False)


310
def _is_1d_list(data: Any) -> bool:
311
    """Check whether data is a 1-D list."""
312
    return isinstance(data, list) and (not data or _is_numeric(data[0]))
wxchan's avatar
wxchan committed
313

wxchan's avatar
wxchan committed
314

315
316
317
318
319
320
321
322
323
324
325
326
327
328
def _is_list_of_numpy_arrays(data: Any) -> "TypeGuard[List[np.ndarray]]":
    return (
        isinstance(data, list)
        and all(isinstance(x, np.ndarray) for x in data)
    )


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


329
330
331
def _is_1d_collection(data: Any) -> bool:
    """Check whether data is a 1-D collection."""
    return (
332
        _is_numpy_1d_array(data)
333
        or _is_numpy_column_array(data)
334
        or _is_1d_list(data)
335
336
337
338
        or isinstance(data, pd_Series)
    )


339
340
def _list_to_1d_numpy(
    data: Any,
341
342
    dtype: "np.typing.DTypeLike",
    name: str
343
) -> np.ndarray:
344
    """Convert data to numpy 1-D array."""
345
    if _is_numpy_1d_array(data):
346
        return _cast_numpy_array_to_dtype(data, dtype)
347
    elif _is_numpy_column_array(data):
348
349
        _log_warning('Converting column-vector to 1d array')
        array = data.ravel()
350
        return _cast_numpy_array_to_dtype(array, dtype)
351
    elif _is_1d_list(data):
wxchan's avatar
wxchan committed
352
        return np.array(data, dtype=dtype, copy=False)
353
    elif isinstance(data, pd_Series):
354
        _check_for_bad_pandas_dtypes(data.to_frame().dtypes)
355
        return np.array(data, dtype=dtype, copy=False)  # SparseArray should be supported as well
wxchan's avatar
wxchan committed
356
    else:
357
358
        raise TypeError(f"Wrong type({type(data).__name__}) for {name}.\n"
                        "It should be list, numpy 1-D array or pandas Series")
wxchan's avatar
wxchan committed
359

wxchan's avatar
wxchan committed
360

361
362
363
364
365
366
367
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."""
368
    return isinstance(data, list) and len(data) > 0 and _is_1d_list(data[0])
369
370
371
372
373
374
375
376
377
378
379


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


380
def _is_pyarrow_array(data: Any) -> "TypeGuard[Union[pa_Array, pa_ChunkedArray]]":
381
382
383
384
    """Check whether data is a PyArrow array."""
    return isinstance(data, (pa_Array, pa_ChunkedArray))


385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
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
416
417
418
419
420
    if isinstance(data, pa_Array):
        export_objects = [data]
    elif isinstance(data, pa_ChunkedArray):
        export_objects = data.chunks
    elif isinstance(data, pa_Table):
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
        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)



442
443
def _data_to_2d_numpy(
    data: Any,
444
445
    dtype: "np.typing.DTypeLike",
    name: str
446
) -> np.ndarray:
447
448
    """Convert data to numpy 2-D array."""
    if _is_numpy_2d_array(data):
449
        return _cast_numpy_array_to_dtype(data, dtype)
450
451
452
    if _is_2d_list(data):
        return np.array(data, dtype=dtype)
    if isinstance(data, pd_DataFrame):
453
        _check_for_bad_pandas_dtypes(data.dtypes)
454
        return _cast_numpy_array_to_dtype(data.values, dtype)
455
456
457
458
    raise TypeError(f"Wrong type({type(data).__name__}) for {name}.\n"
                    "It should be list of lists, numpy 2-D array or pandas DataFrame")


459
def _cfloat32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
460
    """Convert a ctypes float pointer array to a numpy array."""
wxchan's avatar
wxchan committed
461
    if isinstance(cptr, ctypes.POINTER(ctypes.c_float)):
462
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
463
    else:
464
        raise RuntimeError('Expected float pointer')
wxchan's avatar
wxchan committed
465

Guolin Ke's avatar
Guolin Ke committed
466

467
def _cfloat64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
468
    """Convert a ctypes double pointer array to a numpy array."""
Guolin Ke's avatar
Guolin Ke committed
469
    if isinstance(cptr, ctypes.POINTER(ctypes.c_double)):
470
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
Guolin Ke's avatar
Guolin Ke committed
471
472
473
    else:
        raise RuntimeError('Expected double pointer')

wxchan's avatar
wxchan committed
474

475
def _cint32_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
476
    """Convert a ctypes int pointer array to a numpy array."""
wxchan's avatar
wxchan committed
477
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int32)):
478
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
479
    else:
480
481
482
        raise RuntimeError('Expected int32 pointer')


483
def _cint64_array_to_numpy(*, cptr: "ctypes._Pointer", length: int) -> np.ndarray:
484
485
    """Convert a ctypes int pointer array to a numpy array."""
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int64)):
486
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
487
488
    else:
        raise RuntimeError('Expected int64 pointer')
wxchan's avatar
wxchan committed
489

wxchan's avatar
wxchan committed
490

491
def _c_str(string: str) -> ctypes.c_char_p:
492
    """Convert a Python string to C string."""
wxchan's avatar
wxchan committed
493
494
    return ctypes.c_char_p(string.encode('utf-8'))

wxchan's avatar
wxchan committed
495

496
def _c_array(ctype: type, values: List[Any]) -> ctypes.Array:
497
    """Convert a Python array to C array."""
498
    return (ctype * len(values))(*values)  # type: ignore[operator]
wxchan's avatar
wxchan committed
499

wxchan's avatar
wxchan committed
500

501
def _json_default_with_numpy(obj: Any) -> Any:
502
503
504
505
506
507
508
509
510
    """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


511
512
513
514
515
516
517
518
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)


519
def _param_dict_to_str(data: Optional[Dict[str, Any]]) -> str:
520
    """Convert Python dictionary to string, which is passed to C API."""
521
    if data is None or not data:
wxchan's avatar
wxchan committed
522
523
524
        return ""
    pairs = []
    for key, val in data.items():
525
        if isinstance(val, (list, tuple, set)) or _is_numpy_1d_array(val):
526
            pairs.append(f"{key}={','.join(map(_to_string, val))}")
527
        elif isinstance(val, (str, Path, _NUMERIC_TYPES)) or _is_numeric(val):
528
            pairs.append(f"{key}={val}")
529
        elif val is not None:
530
            raise TypeError(f'Unknown type of parameter:{key}, got:{type(val).__name__}')
wxchan's avatar
wxchan committed
531
    return ' '.join(pairs)
532

wxchan's avatar
wxchan committed
533

534
class _TempFile:
535
536
    """Proxy class to workaround errors on Windows."""

537
538
539
    def __enter__(self):
        with NamedTemporaryFile(prefix="lightgbm_tmp_", delete=True) as f:
            self.name = f.name
540
            self.path = Path(self.name)
541
        return self
wxchan's avatar
wxchan committed
542

543
    def __exit__(self, exc_type, exc_val, exc_tb):
544
545
        if self.path.is_file():
            self.path.unlink()
546

wxchan's avatar
wxchan committed
547

548
class LightGBMError(Exception):
549
550
    """Error thrown by LightGBM."""

551
552
553
    pass


554
555
556
557
558
559
560
561
# DeprecationWarning is not shown by default, so let's create our own with higher level
class LGBMDeprecationWarning(UserWarning):
    """Custom deprecation warning."""

    pass


class _ConfigAliases:
562
563
564
565
    # lazy evaluation to allow import without dynamic library, e.g., for docs generation
    aliases = None

    @staticmethod
566
    def _get_all_param_aliases() -> Dict[str, List[str]]:
567
568
569
        buffer_len = 1 << 20
        tmp_out_len = ctypes.c_int64(0)
        string_buffer = ctypes.create_string_buffer(buffer_len)
570
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
571
572
573
574
575
576
577
578
        _safe_call(_LIB.LGBM_DumpParamAliases(
            ctypes.c_int64(buffer_len),
            ctypes.byref(tmp_out_len),
            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)
579
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
580
581
582
583
            _safe_call(_LIB.LGBM_DumpParamAliases(
                ctypes.c_int64(actual_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
584
        return json.loads(
585
            string_buffer.value.decode('utf-8'),
586
            object_hook=lambda obj: {k: [k] + v for k, v in obj.items()}
587
        )
588
589

    @classmethod
590
591
592
    def get(cls, *args) -> Set[str]:
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
593
594
        ret = set()
        for i in args:
595
            ret.update(cls.get_sorted(i))
596
597
        return ret

598
599
600
601
602
603
    @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])

604
    @classmethod
605
606
607
    def get_by_alias(cls, *args) -> Set[str]:
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
608
609
610
611
        ret = set(args)
        for arg in args:
            for aliases in cls.aliases.values():
                if arg in aliases:
612
                    ret.update(aliases)
613
614
615
                    break
        return ret

616

617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
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)

638
639
    aliases = _ConfigAliases.get_sorted(main_param_name)
    aliases = [a for a in aliases if a != main_param_name]
640
641

    # if main_param_name was provided, keep that value and remove all aliases
642
    if main_param_name in params.keys():
643
644
645
        for param in aliases:
            params.pop(param, None)
        return params
646

647
648
649
650
651
    # 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
652

653
654
655
656
657
658
659
    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
660
661
662
663

    return params


664
_MAX_INT32 = (1 << 31) - 1
665

666
"""Macro definition of data type in C API of LightGBM"""
667
668
669
670
_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
671

672
"""Matrix is row major in Python"""
673
_C_API_IS_ROW_MAJOR = 1
wxchan's avatar
wxchan committed
674

675
"""Macro definition of prediction type in C API of LightGBM"""
676
677
678
679
_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
680

681
"""Macro definition of sparse matrix type"""
682
683
_C_API_MATRIX_TYPE_CSR = 0
_C_API_MATRIX_TYPE_CSC = 1
684

685
"""Macro definition of feature importance type"""
686
687
_C_API_FEATURE_IMPORTANCE_SPLIT = 0
_C_API_FEATURE_IMPORTANCE_GAIN = 1
688

689
"""Data type of data field"""
690
691
692
693
_FIELD_TYPE_MAPPER = {
    "label": _C_API_DTYPE_FLOAT32,
    "weight": _C_API_DTYPE_FLOAT32,
    "init_score": _C_API_DTYPE_FLOAT64,
694
695
    "group": _C_API_DTYPE_INT32,
    "position": _C_API_DTYPE_INT32
696
}
wxchan's avatar
wxchan committed
697

698
"""String name to int feature importance type mapper"""
699
700
701
702
_FEATURE_IMPORTANCE_TYPE_MAPPER = {
    "split": _C_API_FEATURE_IMPORTANCE_SPLIT,
    "gain": _C_API_FEATURE_IMPORTANCE_GAIN
}
703

wxchan's avatar
wxchan committed
704

705
def _convert_from_sliced_object(data: np.ndarray) -> np.ndarray:
706
    """Fix the memory of multi-dimensional sliced object."""
707
    if isinstance(data, np.ndarray) and isinstance(data.base, np.ndarray):
708
        if not data.flags.c_contiguous:
709
710
            _log_warning("Usage of np.ndarray subset (sliced data) is not recommended "
                         "due to it will double the peak memory cost in LightGBM.")
711
712
713
714
            return np.copy(data)
    return data


715
716
717
def _c_float_array(
    data: np.ndarray
) -> Tuple[_ctypes_float_ptr, int, np.ndarray]:
718
    """Get pointer of float numpy array / list."""
719
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
720
        data = np.array(data, copy=False)
721
    if _is_numpy_1d_array(data):
722
        data = _convert_from_sliced_object(data)
723
        assert data.flags.c_contiguous
724
        ptr_data: _ctypes_float_ptr
wxchan's avatar
wxchan committed
725
726
        if data.dtype == np.float32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
727
            type_data = _C_API_DTYPE_FLOAT32
wxchan's avatar
wxchan committed
728
729
        elif data.dtype == np.float64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_double))
730
            type_data = _C_API_DTYPE_FLOAT64
wxchan's avatar
wxchan committed
731
        else:
732
            raise TypeError(f"Expected np.float32 or np.float64, met type({data.dtype})")
wxchan's avatar
wxchan committed
733
    else:
734
        raise TypeError(f"Unknown type({type(data).__name__})")
735
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
736

wxchan's avatar
wxchan committed
737

738
739
740
def _c_int_array(
    data: np.ndarray
) -> Tuple[_ctypes_int_ptr, int, np.ndarray]:
741
    """Get pointer of int numpy array / list."""
742
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
743
        data = np.array(data, copy=False)
744
    if _is_numpy_1d_array(data):
745
        data = _convert_from_sliced_object(data)
746
        assert data.flags.c_contiguous
747
        ptr_data: _ctypes_int_ptr
wxchan's avatar
wxchan committed
748
749
        if data.dtype == np.int32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int32))
750
            type_data = _C_API_DTYPE_INT32
wxchan's avatar
wxchan committed
751
752
        elif data.dtype == np.int64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int64))
753
            type_data = _C_API_DTYPE_INT64
wxchan's avatar
wxchan committed
754
        else:
755
            raise TypeError(f"Expected np.int32 or np.int64, met type({data.dtype})")
wxchan's avatar
wxchan committed
756
    else:
757
        raise TypeError(f"Unknown type({type(data).__name__})")
758
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
759

wxchan's avatar
wxchan committed
760

761
def _is_allowed_numpy_dtype(dtype: type) -> bool:
762
    float128 = getattr(np, 'float128', type(None))
763
764
765
766
    return (
        issubclass(dtype, (np.integer, np.floating, np.bool_))
        and not issubclass(dtype, (np.timedelta64, float128))
    )
767
768


769
def _check_for_bad_pandas_dtypes(pandas_dtypes_series: pd_Series) -> None:
770
771
    bad_pandas_dtypes = [
        f'{column_name}: {pandas_dtype}'
772
        for column_name, pandas_dtype in pandas_dtypes_series.items()
773
        if not _is_allowed_numpy_dtype(pandas_dtype.type)
774
775
776
777
    ]
    if bad_pandas_dtypes:
        raise ValueError('pandas dtypes must be int, float or bool.\n'
                         f'Fields with bad pandas dtypes: {", ".join(bad_pandas_dtypes)}')
778
779


780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
def _pandas_to_numpy(
    data: pd_DataFrame,
    target_dtype: "np.typing.DTypeLike"
) -> 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)


797
def _data_from_pandas(
798
799
800
    data: pd_DataFrame,
    feature_name: _LGBM_FeatureNameConfiguration,
    categorical_feature: _LGBM_CategoricalFeatureConfiguration,
801
    pandas_categorical: Optional[List[List]]
802
) -> Tuple[np.ndarray, List[str], Union[List[str], List[int]], List[List]]:
803
804
805
    if len(data.shape) != 2 or data.shape[0] < 1:
        raise ValueError('Input data must be 2 dimensional and non empty.')

806
807
808
809
    # take shallow copy in case we modify categorical columns
    # whole column modifications don't change the original df
    data = data.copy(deep=False)

810
811
812
813
814
815
    # determine feature names
    if feature_name == 'auto':
        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)]
816
    cat_cols_not_ordered: List[str] = [col for col in cat_cols if not data[col].cat.ordered]
817
818
    if pandas_categorical is None:  # train dataset
        pandas_categorical = [list(data[col].cat.categories) for col in cat_cols]
819
    else:
820
821
822
823
824
825
826
        if len(cat_cols) != len(pandas_categorical):
            raise ValueError('train and valid dataset categorical_feature do not match.')
        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})
827
828
829

    # use cat cols from DataFrame
    if categorical_feature == 'auto':
830
831
832
        categorical_feature = cat_cols_not_ordered

    df_dtypes = [dtype.type for dtype in data.dtypes]
833
834
    # so that the target dtype considers floats
    df_dtypes.append(np.float32)
835
    target_dtype = np.result_type(*df_dtypes)
836
837
838
839
840
841
842

    return (
        _pandas_to_numpy(data, target_dtype=target_dtype),
        feature_name,
        categorical_feature,
        pandas_categorical
    )
843
844


845
846
847
848
def _dump_pandas_categorical(
    pandas_categorical: Optional[List[List]],
    file_name: Optional[Union[str, Path]] = None
) -> str:
849
    categorical_json = json.dumps(pandas_categorical, default=_json_default_with_numpy)
850
    pandas_str = f'\npandas_categorical:{categorical_json}\n'
851
852
853
854
855
856
    if file_name is not None:
        with open(file_name, 'a') as f:
            f.write(pandas_str)
    return pandas_str


857
858
859
def _load_pandas_categorical(
    file_name: Optional[Union[str, Path]] = None,
    model_str: Optional[str] = None
860
) -> Optional[List[List]]:
861
862
    pandas_key = 'pandas_categorical:'
    offset = -len(pandas_key)
863
    if file_name is not None:
864
        max_offset = -getsize(file_name)
865
866
867
868
        with open(file_name, 'rb') as f:
            while True:
                if offset < max_offset:
                    offset = max_offset
869
                f.seek(offset, SEEK_END)
870
871
872
873
                lines = f.readlines()
                if len(lines) >= 2:
                    break
                offset *= 2
874
        last_line = lines[-1].decode('utf-8').strip()
875
        if not last_line.startswith(pandas_key):
876
            last_line = lines[-2].decode('utf-8').strip()
877
    elif model_str is not None:
878
879
880
881
882
883
        idx = model_str.rfind('\n', 0, offset)
        last_line = model_str[idx:].strip()
    if last_line.startswith(pandas_key):
        return json.loads(last_line[len(pandas_key):])
    else:
        return None
884
885


886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
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**.

906
907
    .. versionadded:: 3.3.0

908
909
910
911
912
913
914
915
916
    Attributes
    ----------
    batch_size : int
        Default size of a batch.
    """

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

    @abc.abstractmethod
917
    def __getitem__(self, idx: Union[int, slice, List[int]]) -> np.ndarray:
918
919
920
921
922
923
924
        """Return data for given row index.

        A basic implementation should look like this:

        .. code-block:: python

            if isinstance(idx, numbers.Integral):
925
                return self._get_one_line(idx)
926
            elif isinstance(idx, slice):
927
928
                return np.stack([self._get_one_line(i) for i in range(idx.start, idx.stop)])
            elif isinstance(idx, list):
929
                # Only required if using ``Dataset.subset()``.
930
                return np.array([self._get_one_line(i) for i in idx])
931
            else:
932
                raise TypeError(f"Sequence index must be integer, slice or list, got {type(idx).__name__}")
933
934
935

        Parameters
        ----------
936
        idx : int, slice[int], list[int]
937
938
939
940
            Item index.

        Returns
        -------
941
        result : numpy 1-D array or numpy 2-D array
942
            1-D array if idx is int, 2-D array if idx is slice or list.
943
944
945
946
947
948
949
950
951
        """
        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__()")


952
class _InnerPredictor:
953
954
955
956
957
    """_InnerPredictor of LightGBM.

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

Nikita Titov's avatar
Nikita Titov committed
958
959
960
    .. note::

        Can be converted from Booster, but cannot be converted to Booster.
Guolin Ke's avatar
Guolin Ke committed
961
    """
962

963
964
    def __init__(
        self,
965
966
967
968
        booster_handle: _BoosterHandle,
        pandas_categorical: Optional[List[List]],
        pred_parameter: Dict[str, Any],
        manage_handle: bool
969
    ):
970
        """Initialize the _InnerPredictor.
wxchan's avatar
wxchan committed
971
972
973

        Parameters
        ----------
974
        booster_handle : object
975
            Handle of Booster.
976
977
978
979
        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
980
            Other parameters for the prediction.
981
982
        manage_handle : bool
            If ``True``, free the corresponding Booster on the C++ side when this Python object is deleted.
wxchan's avatar
wxchan committed
983
        """
984
985
986
987
988
989
990
991
        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(
992
                self._handle,
993
994
995
996
                ctypes.byref(out_num_class)
            )
        )
        self.num_class = out_num_class.value
wxchan's avatar
wxchan committed
997

998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
    @classmethod
    def from_booster(
        cls,
        booster: "Booster",
        pred_parameter: Dict[str, Any]
    ) -> "_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,
                ctypes.byref(out_cur_iter)
            )
        )
        return cls(
            booster_handle=booster._handle,
            pandas_categorical=booster.pandas_categorical,
            pred_parameter=pred_parameter,
            manage_handle=False
        )

    @classmethod
    def from_model_file(
        cls,
        model_file: Union[str, Path],
        pred_parameter: Dict[str, Any]
    ) -> "_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),
                ctypes.byref(booster_handle)
            )
        )
        return cls(
            booster_handle=booster_handle,
            pandas_categorical=_load_pandas_categorical(file_name=model_file),
            pred_parameter=pred_parameter,
            manage_handle=True
        )
cbecker's avatar
cbecker committed
1057

1058
    def __del__(self) -> None:
1059
1060
        try:
            if self.__is_manage_handle:
1061
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
1062
1063
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
1064

1065
    def __getstate__(self) -> Dict[str, Any]:
1066
1067
        this = self.__dict__.copy()
        this.pop('handle', None)
1068
        this.pop('_handle', None)
1069
1070
        return this

1071
1072
    def predict(
        self,
1073
        data: _LGBM_PredictDataType,
1074
1075
1076
1077
1078
1079
1080
        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,
        validate_features: bool = False
1081
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
1082
        """Predict logic.
wxchan's avatar
wxchan committed
1083
1084
1085

        Parameters
        ----------
1086
        data : str, pathlib.Path, numpy array, pandas DataFrame, pyarrow Table, H2O DataTable's Frame or scipy.sparse
1087
            Data source for prediction.
1088
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
1089
1090
        start_iteration : int, optional (default=0)
            Start index of the iteration to predict.
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
        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.
1102
1103
1104
        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
1105

1106
1107
            .. versionadded:: 4.0.0

wxchan's avatar
wxchan committed
1108
1109
        Returns
        -------
1110
        result : numpy array, scipy.sparse or list of scipy.sparse
1111
            Prediction result.
1112
            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
1113
        """
wxchan's avatar
wxchan committed
1114
        if isinstance(data, Dataset):
1115
            raise TypeError("Cannot use Dataset instance for prediction, please use raw data instead")
1116
1117
1118
1119
1120
1121
        elif isinstance(data, pd_DataFrame) and validate_features:
            data_names = [str(x) for x in data.columns]
            ptr_names = (ctypes.c_char_p * len(data_names))()
            ptr_names[:] = [x.encode('utf-8') for x in data_names]
            _safe_call(
                _LIB.LGBM_BoosterValidateFeatureNames(
1122
                    self._handle,
1123
1124
1125
1126
                    ptr_names,
                    ctypes.c_int(len(data_names)),
                )
            )
1127
1128
1129
1130
1131
1132
1133
1134
1135

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

1136
        predict_type = _C_API_PREDICT_NORMAL
wxchan's avatar
wxchan committed
1137
        if raw_score:
1138
            predict_type = _C_API_PREDICT_RAW_SCORE
wxchan's avatar
wxchan committed
1139
        if pred_leaf:
1140
            predict_type = _C_API_PREDICT_LEAF_INDEX
1141
        if pred_contrib:
1142
            predict_type = _C_API_PREDICT_CONTRIB
cbecker's avatar
cbecker committed
1143

1144
        if isinstance(data, (str, Path)):
1145
            with _TempFile() as f:
wxchan's avatar
wxchan committed
1146
                _safe_call(_LIB.LGBM_BoosterPredictForFile(
1147
                    self._handle,
1148
                    _c_str(str(data)),
1149
                    ctypes.c_int(data_has_header),
Guolin Ke's avatar
Guolin Ke committed
1150
                    ctypes.c_int(predict_type),
1151
                    ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
1152
                    ctypes.c_int(num_iteration),
1153
1154
                    _c_str(self.pred_parameter),
                    _c_str(f.name)))
1155
1156
                preds = np.loadtxt(f.name, dtype=np.float64)
                nrow = preds.shape[0]
wxchan's avatar
wxchan committed
1157
        elif isinstance(data, scipy.sparse.csr_matrix):
1158
1159
1160
1161
1162
1163
            preds, nrow = self.__pred_for_csr(
                csr=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
Guolin Ke's avatar
Guolin Ke committed
1164
        elif isinstance(data, scipy.sparse.csc_matrix):
1165
1166
1167
1168
1169
1170
            preds, nrow = self.__pred_for_csc(
                csc=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
wxchan's avatar
wxchan committed
1171
        elif isinstance(data, np.ndarray):
1172
1173
1174
1175
1176
1177
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1178
1179
1180
1181
1182
1183
1184
        elif _is_pyarrow_table(data):
            preds, nrow = self.__pred_for_pyarrow_table(
                table=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1185
1186
1187
        elif isinstance(data, list):
            try:
                data = np.array(data)
1188
1189
            except BaseException as err:
                raise ValueError('Cannot convert data list to numpy array.') from err
1190
1191
1192
1193
1194
1195
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1196
        elif isinstance(data, dt_DataTable):
1197
1198
1199
1200
1201
1202
            preds, nrow = self.__pred_for_np2d(
                mat=data.to_numpy(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
wxchan's avatar
wxchan committed
1203
1204
        else:
            try:
1205
                _log_warning('Converting data to scipy sparse matrix.')
wxchan's avatar
wxchan committed
1206
                csr = scipy.sparse.csr_matrix(data)
1207
1208
            except BaseException as err:
                raise TypeError(f'Cannot predict data for type {type(data).__name__}') from err
1209
1210
1211
1212
1213
1214
            preds, nrow = self.__pred_for_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
wxchan's avatar
wxchan committed
1215
1216
        if pred_leaf:
            preds = preds.astype(np.int32)
1217
        is_sparse = isinstance(preds, (list, scipy.sparse.spmatrix))
1218
        if not is_sparse and preds.size != nrow:
wxchan's avatar
wxchan committed
1219
            if preds.size % nrow == 0:
1220
                preds = preds.reshape(nrow, -1)
wxchan's avatar
wxchan committed
1221
            else:
1222
                raise ValueError(f'Length of predict result ({preds.size}) cannot be divide nrow ({nrow})')
wxchan's avatar
wxchan committed
1223
1224
        return preds

1225
1226
1227
1228
1229
1230
1231
    def __get_num_preds(
        self,
        start_iteration: int,
        num_iteration: int,
        nrow: int,
        predict_type: int
    ) -> int:
1232
        """Get size of prediction result."""
1233
        if nrow > _MAX_INT32:
1234
            raise LightGBMError('LightGBM cannot perform prediction for data '
1235
                                f'with number of rows greater than MAX_INT32 ({_MAX_INT32}).\n'
1236
                                'You can split your data into chunks '
1237
                                'and then concatenate predictions for them')
Guolin Ke's avatar
Guolin Ke committed
1238
1239
        n_preds = ctypes.c_int64(0)
        _safe_call(_LIB.LGBM_BoosterCalcNumPredict(
1240
            self._handle,
Guolin Ke's avatar
Guolin Ke committed
1241
1242
            ctypes.c_int(nrow),
            ctypes.c_int(predict_type),
1243
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
1244
            ctypes.c_int(num_iteration),
Guolin Ke's avatar
Guolin Ke committed
1245
1246
            ctypes.byref(n_preds)))
        return n_preds.value
wxchan's avatar
wxchan committed
1247

1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
    def __inner_predict_np2d(
        self,
        mat: np.ndarray,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
        preds: Optional[np.ndarray]
    ) -> Tuple[np.ndarray, int]:
        if mat.dtype == np.float32 or mat.dtype == np.float64:
            data = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
        else:  # change non-float data to float data, need to copy
            data = np.array(mat.reshape(mat.size), dtype=np.float32)
        ptr_data, type_ptr_data, _ = _c_float_array(data)
1261
1262
1263
1264
1265
1266
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=mat.shape[0],
            predict_type=predict_type
        )
1267
1268
1269
1270
1271
1272
        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)
        _safe_call(_LIB.LGBM_BoosterPredictForMat(
1273
            self._handle,
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
            ptr_data,
            ctypes.c_int(type_ptr_data),
            ctypes.c_int32(mat.shape[0]),
            ctypes.c_int32(mat.shape[1]),
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
            ctypes.c_int(predict_type),
            ctypes.c_int(start_iteration),
            ctypes.c_int(num_iteration),
            _c_str(self.pred_parameter),
            ctypes.byref(out_num_preds),
            preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
        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,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
1296
        """Predict for a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
1297
        if len(mat.shape) != 2:
1298
            raise ValueError('Input numpy.ndarray or list must be 2 dimensional')
wxchan's avatar
wxchan committed
1299

1300
        nrow = mat.shape[0]
1301
1302
        if nrow > _MAX_INT32:
            sections = np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)
1303
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1304
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff([0] + list(sections) + [nrow])]
1305
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1306
            preds = np.empty(sum(n_preds), dtype=np.float64)
1307
1308
            for chunk, (start_idx_pred, end_idx_pred) in zip(np.array_split(mat, sections),
                                                             zip(n_preds_sections, n_preds_sections[1:])):
1309
                # avoid memory consumption by arrays concatenation operations
1310
1311
1312
1313
1314
1315
1316
                self.__inner_predict_np2d(
                    mat=chunk,
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
                    preds=preds[start_idx_pred:end_idx_pred]
                )
1317
            return preds, nrow
wxchan's avatar
wxchan committed
1318
        else:
1319
1320
1321
1322
1323
1324
1325
            return self.__inner_predict_np2d(
                mat=mat,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
                preds=None
            )
wxchan's avatar
wxchan committed
1326

1327
1328
1329
    def __create_sparse_native(
        self,
        cs: Union[scipy.sparse.csc_matrix, scipy.sparse.csr_matrix],
1330
1331
1332
1333
1334
1335
        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,
1336
        is_csr: bool
1337
    ) -> Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]]:
1338
1339
1340
        # create numpy array from output arrays
        data_indices_len = out_shape[0]
        indptr_len = out_shape[1]
1341
        if indptr_type == _C_API_DTYPE_INT32:
1342
            out_indptr = _cint32_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1343
        elif indptr_type == _C_API_DTYPE_INT64:
1344
            out_indptr = _cint64_array_to_numpy(cptr=out_ptr_indptr, length=indptr_len)
1345
1346
        else:
            raise TypeError("Expected int32 or int64 type for indptr")
1347
        if data_type == _C_API_DTYPE_FLOAT32:
1348
            out_data = _cfloat32_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1349
        elif data_type == _C_API_DTYPE_FLOAT64:
1350
            out_data = _cfloat64_array_to_numpy(cptr=out_ptr_data, length=data_indices_len)
1351
1352
        else:
            raise TypeError("Expected float32 or float64 type for data")
1353
        out_indices = _cint32_array_to_numpy(cptr=out_ptr_indices, length=data_indices_len)
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
        # 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]
            cs_indices = out_indices[offset + cs_indptr[0]:offset + matrix_indptr_len]
            cs_data = out_data[offset + cs_indptr[0]:offset + matrix_indptr_len]
            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
        _safe_call(_LIB.LGBM_BoosterFreePredictSparse(out_ptr_indptr, out_ptr_indices, out_ptr_data,
                                                      ctypes.c_int(indptr_type), ctypes.c_int(data_type)))
        if len(cs_output_matrices) == 1:
            return cs_output_matrices[0]
        return cs_output_matrices

1382
1383
1384
1385
1386
1387
1388
1389
1390
    def __inner_predict_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int,
        preds: Optional[np.ndarray]
    ) -> Tuple[np.ndarray, int]:
        nrow = len(csr.indptr) - 1
1391
1392
1393
1394
1395
1396
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
            predict_type=predict_type
        )
1397
1398
1399
1400
1401
        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
1402

1403
1404
        ptr_indptr, type_ptr_indptr, _ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
1405

1406
1407
1408
1409
        assert csr.shape[1] <= _MAX_INT32
        csr_indices = csr.indices.astype(np.int32, copy=False)

        _safe_call(_LIB.LGBM_BoosterPredictForCSR(
1410
            self._handle,
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
            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))))
        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,
        predict_type: int
1435
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1436
1437
1438
1439
        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
1440
        out_ptr_indptr: _ctypes_int_ptr
1441
1442
1443
1444
1445
        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)()
1446
        out_ptr_data: _ctypes_float_ptr
1447
1448
1449
1450
1451
1452
        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)
        _safe_call(_LIB.LGBM_BoosterPredictSparseOutput(
1453
            self._handle,
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
            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)))
        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,
            is_csr=True
        )
        nrow = len(csr.indptr) - 1
        return matrices, nrow

1484
1485
1486
1487
1488
1489
1490
    def __pred_for_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
1491
        """Predict for a CSR data."""
1492
        if predict_type == _C_API_PREDICT_CONTRIB:
1493
1494
1495
1496
1497
1498
            return self.__inner_predict_csr_sparse(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1499
        nrow = len(csr.indptr) - 1
1500
1501
        if nrow > _MAX_INT32:
            sections = [0] + list(np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)) + [nrow]
1502
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1503
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff(sections)]
1504
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1505
            preds = np.empty(sum(n_preds), dtype=np.float64)
1506
1507
            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:])):
1508
                # avoid memory consumption by arrays concatenation operations
1509
1510
1511
1512
1513
1514
1515
                self.__inner_predict_csr(
                    csr=csr[start_idx:end_idx],
                    start_iteration=start_iteration,
                    num_iteration=num_iteration,
                    predict_type=predict_type,
                    preds=preds[start_idx_pred:end_idx_pred]
                )
1516
1517
            return preds, nrow
        else:
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
            return self.__inner_predict_csr(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type,
                preds=None
            )

    def __inner_predict_sparse_csc(
        self,
1528
1529
1530
1531
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
1532
1533
1534
1535
1536
    ):
        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
1537
        out_ptr_indptr: _ctypes_int_ptr
1538
1539
1540
1541
1542
        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)()
1543
        out_ptr_data: _ctypes_float_ptr
1544
1545
1546
1547
1548
1549
        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)
        _safe_call(_LIB.LGBM_BoosterPredictSparseOutput(
1550
            self._handle,
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
            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)))
        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,
            is_csr=False
        )
        nrow = csc.shape[0]
        return matrices, nrow
Guolin Ke's avatar
Guolin Ke committed
1580

1581
1582
1583
1584
1585
1586
1587
    def __pred_for_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
1588
        """Predict for a CSC data."""
Guolin Ke's avatar
Guolin Ke committed
1589
        nrow = csc.shape[0]
1590
        if nrow > _MAX_INT32:
1591
1592
1593
1594
1595
1596
            return self.__pred_for_csr(
                csr=csc.tocsr(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1597
        if predict_type == _C_API_PREDICT_CONTRIB:
1598
1599
1600
1601
1602
1603
            return self.__inner_predict_sparse_csc(
                csc=csc,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1604
1605
1606
1607
1608
1609
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
            predict_type=predict_type
        )
1610
        preds = np.empty(n_preds, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
1611
1612
        out_num_preds = ctypes.c_int64(0)

1613
1614
        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
1615

1616
        assert csc.shape[0] <= _MAX_INT32
1617
        csc_indices = csc.indices.astype(np.int32, copy=False)
1618

Guolin Ke's avatar
Guolin Ke committed
1619
        _safe_call(_LIB.LGBM_BoosterPredictForCSC(
1620
            self._handle,
Guolin Ke's avatar
Guolin Ke committed
1621
            ptr_indptr,
1622
            ctypes.c_int(type_ptr_indptr),
1623
            csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
Guolin Ke's avatar
Guolin Ke committed
1624
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
1625
1626
1627
1628
1629
            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),
1630
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
1631
            ctypes.c_int(num_iteration),
1632
            _c_str(self.pred_parameter),
Guolin Ke's avatar
Guolin Ke committed
1633
            ctypes.byref(out_num_preds),
wxchan's avatar
wxchan committed
1634
            preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
wxchan's avatar
wxchan committed
1635
        if n_preds != out_num_preds.value:
1636
            raise ValueError("Wrong length for predict results")
wxchan's avatar
wxchan committed
1637
        return preds, nrow
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
    
    def __pred_for_pyarrow_table(
        self,
        table: pa_Table,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
        """Predict for a PyArrow table."""
        if not PYARROW_INSTALLED:
            raise LightGBMError("Cannot predict from Arrow without `pyarrow` installed.")

        # Check that the input is valid: we only handle numbers (for now)
        if not all(arrow_is_integer(t) or arrow_is_floating(t) for t in table.schema.types):
            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,
            predict_type=predict_type
        )
        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)
        _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))))
        if n_preds != out_num_preds.value:
            raise ValueError("Wrong length for predict results")
        return preds, table.num_rows
wxchan's avatar
wxchan committed
1680

1681
    def current_iteration(self) -> int:
1682
1683
1684
1685
1686
1687
1688
1689
1690
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
        out_cur_iter = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterGetCurrentIteration(
1691
            self._handle,
1692
1693
1694
            ctypes.byref(out_cur_iter)))
        return out_cur_iter.value

wxchan's avatar
wxchan committed
1695

1696
class Dataset:
wxchan's avatar
wxchan committed
1697
    """Dataset in LightGBM."""
1698

1699
1700
    def __init__(
        self,
1701
        data: _LGBM_TrainDataType,
1702
        label: Optional[_LGBM_LabelType] = None,
1703
        reference: Optional["Dataset"] = None,
1704
1705
1706
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
1707
1708
        feature_name: _LGBM_FeatureNameConfiguration = 'auto',
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = 'auto',
1709
        params: Optional[Dict[str, Any]] = None,
1710
1711
        free_raw_data: bool = True,
        position: Optional[_LGBM_PositionType] = None,
1712
    ):
1713
        """Initialize Dataset.
1714

wxchan's avatar
wxchan committed
1715
1716
        Parameters
        ----------
1717
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence, list of numpy array or pyarrow Table
wxchan's avatar
wxchan committed
1718
            Data source of Dataset.
1719
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
1720
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1721
1722
1723
            Label of the data.
        reference : Dataset or None, optional (default=None)
            If this is Dataset for validation, training data should be used as reference.
1724
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1725
            Weight for each instance. Weights should be non-negative.
1726
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
1727
1728
1729
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
1730
1731
            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.
1732
        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)
1733
            Init score for Dataset.
1734
        feature_name : list of str, or 'auto', optional (default="auto")
1735
            Feature names.
1736
            If 'auto' and data is pandas DataFrame or pyarrow Table, data columns names are used.
1737
        categorical_feature : list of str or int, or 'auto', optional (default="auto")
1738
1739
            Categorical features.
            If list of int, interpreted as indices.
1740
            If list of str, interpreted as feature names (need to specify ``feature_name`` as well).
1741
            If 'auto' and data is pandas DataFrame, pandas unordered categorical columns are used.
1742
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
1743
            Large values could be memory consuming. Consider using consecutive integers starting from zero.
1744
            All negative values in categorical features will be treated as missing values.
1745
            The output cannot be monotonically constrained with respect to a categorical feature.
1746
            Floating point numbers in categorical features will be rounded towards 0.
Nikita Titov's avatar
Nikita Titov committed
1747
        params : dict or None, optional (default=None)
1748
            Other parameters for Dataset.
Nikita Titov's avatar
Nikita Titov committed
1749
        free_raw_data : bool, optional (default=True)
1750
            If True, raw data is freed after constructing inner Dataset.
1751
1752
        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
1753
        """
1754
        self._handle: Optional[_DatasetHandle] = None
wxchan's avatar
wxchan committed
1755
1756
1757
1758
1759
        self.data = data
        self.label = label
        self.reference = reference
        self.weight = weight
        self.group = group
1760
        self.position = position
1761
        self.init_score = init_score
1762
1763
        self.feature_name: _LGBM_FeatureNameConfiguration = feature_name
        self.categorical_feature: _LGBM_CategoricalFeatureConfiguration = categorical_feature
1764
        self.params = deepcopy(params)
wxchan's avatar
wxchan committed
1765
        self.free_raw_data = free_raw_data
1766
        self.used_indices: Optional[List[int]] = None
1767
        self._need_slice = True
1768
        self._predictor: Optional[_InnerPredictor] = None
1769
        self.pandas_categorical: Optional[List[List]] = None
1770
        self._params_back_up = None
1771
        self.version = 0
1772
        self._start_row = 0  # Used when pushing rows one by one.
wxchan's avatar
wxchan committed
1773

1774
    def __del__(self) -> None:
1775
1776
1777
1778
        try:
            self._free_handle()
        except AttributeError:
            pass
1779

1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
    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.
        """
1797
        param_str = _param_dict_to_str(self.get_params())
1798
1799
        sample_cnt = _get_sample_count(total_nrow, param_str)
        indices = np.empty(sample_cnt, dtype=np.int32)
1800
        ptr_data, _, _ = _c_int_array(indices)
1801
1802
1803
1804
        actual_sample_cnt = ctypes.c_int32(0)

        _safe_call(_LIB.LGBM_SampleIndices(
            ctypes.c_int32(total_nrow),
1805
            _c_str(param_str),
1806
1807
1808
            ptr_data,
            ctypes.byref(actual_sample_cnt),
        ))
1809
1810
        assert sample_cnt == actual_sample_cnt.value
        return indices
1811

1812
1813
1814
1815
1816
    def _init_from_ref_dataset(
        self,
        total_nrow: int,
        ref_dataset: _DatasetHandle
    ) -> 'Dataset':
1817
1818
1819
1820
1821
1822
        """Create dataset from a reference dataset.

        Parameters
        ----------
        total_nrow : int
            Number of rows expected to add to dataset.
1823
1824
        ref_dataset : object
            Handle of reference dataset to extract metadata from.
1825
1826
1827
1828
1829
1830

        Returns
        -------
        self : Dataset
            Constructed Dataset object.
        """
1831
        self._handle = ctypes.c_void_p()
1832
1833
1834
        _safe_call(_LIB.LGBM_DatasetCreateByReference(
            ref_dataset,
            ctypes.c_int64(total_nrow),
1835
            ctypes.byref(self._handle),
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
        ))
        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
        ----------
1850
        sample_data : list of numpy array
1851
            Sample data for each column.
1852
        sample_indices : list of numpy array
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
            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.
1875
        sample_col_ptr: _ctypes_float_array = (ctypes.POINTER(ctypes.c_double) * ncol)()
1876
1877
        # c type int**
        # each int* points to start of indices for each column
1878
        indices_col_ptr: _ctypes_int_array = (ctypes.POINTER(ctypes.c_int32) * ncol)()
1879
        for i in range(ncol):
1880
1881
            sample_col_ptr[i] = _c_float_array(sample_data[i])[0]
            indices_col_ptr[i] = _c_int_array(sample_indices[i])[0]
1882
1883

        num_per_col = np.array([len(d) for d in sample_indices], dtype=np.int32)
1884
        num_per_col_ptr, _, _ = _c_int_array(num_per_col)
1885

1886
        self._handle = ctypes.c_void_p()
1887
        params_str = _param_dict_to_str(self.get_params())
1888
1889
1890
1891
1892
1893
1894
        _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),
1895
            ctypes.c_int64(total_nrow),
1896
            _c_str(params_str),
1897
            ctypes.byref(self._handle),
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
        ))
        return self

    def _push_rows(self, data: np.ndarray) -> 'Dataset':
        """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)
1916
        data_ptr, data_type, _ = _c_float_array(data)
1917
1918

        _safe_call(_LIB.LGBM_DatasetPushRows(
1919
            self._handle,
1920
1921
1922
1923
1924
1925
1926
1927
1928
            data_ptr,
            data_type,
            ctypes.c_int32(nrow),
            ctypes.c_int32(ncol),
            ctypes.c_int32(self._start_row),
        ))
        self._start_row += nrow
        return self

1929
    def get_params(self) -> Dict[str, Any]:
1930
1931
1932
1933
        """Get the used parameters in the Dataset.

        Returns
        -------
1934
        params : dict
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
            The used parameters in this Dataset object.
        """
        if self.params is not None:
            # no min_data, nthreads and verbose in this function
            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",
1950
                                                "linear_tree",
1951
1952
1953
1954
                                                "max_bin",
                                                "max_bin_by_feature",
                                                "min_data_in_bin",
                                                "pre_partition",
Nikita Titov's avatar
Nikita Titov committed
1955
                                                "precise_float_parser",
1956
1957
1958
1959
1960
                                                "two_round",
                                                "use_missing",
                                                "weight_column",
                                                "zero_as_missing")
            return {k: v for k, v in self.params.items() if k in dataset_params}
1961
1962
        else:
            return {}
1963

1964
    def _free_handle(self) -> "Dataset":
1965
1966
1967
        if self._handle is not None:
            _safe_call(_LIB.LGBM_DatasetFree(self._handle))
            self._handle = None
1968
        self._need_slice = True
Guolin Ke's avatar
Guolin Ke committed
1969
1970
        if self.used_indices is not None:
            self.data = None
Nikita Titov's avatar
Nikita Titov committed
1971
        return self
wxchan's avatar
wxchan committed
1972

1973
1974
1975
    def _set_init_score_by_predictor(
        self,
        predictor: Optional[_InnerPredictor],
1976
        data: _LGBM_TrainDataType,
1977
        used_indices: Optional[Union[List[int], np.ndarray]]
1978
    ) -> "Dataset":
Guolin Ke's avatar
Guolin Ke committed
1979
        data_has_header = False
1980
        if isinstance(data, (str, Path)) and self.params is not None:
Guolin Ke's avatar
Guolin Ke committed
1981
            # check data has header or not
1982
            data_has_header = any(self.params.get(alias, False) for alias in _ConfigAliases.get("header"))
Guolin Ke's avatar
Guolin Ke committed
1983
        num_data = self.num_data()
1984
        if predictor is not None:
1985
1986
1987
1988
1989
            init_score: Union[np.ndarray, scipy.sparse.spmatrix] = predictor.predict(
                data=data,
                raw_score=True,
                data_has_header=data_has_header
            )
1990
            init_score = init_score.ravel()
1991
            if used_indices is not None:
1992
                assert not self._need_slice
1993
                if isinstance(data, (str, Path)):
1994
                    sub_init_score = np.empty(num_data * predictor.num_class, dtype=np.float64)
1995
                    assert num_data == len(used_indices)
1996
1997
                    for i in range(len(used_indices)):
                        for j in range(predictor.num_class):
1998
1999
2000
2001
                            sub_init_score[i * predictor.num_class + j] = init_score[used_indices[i] * predictor.num_class + j]
                    init_score = sub_init_score
            if predictor.num_class > 1:
                # need to regroup init_score
2002
                new_init_score = np.empty(init_score.size, dtype=np.float64)
2003
2004
                for i in range(num_data):
                    for j in range(predictor.num_class):
2005
2006
2007
                        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:
2008
            init_score = np.full_like(self.init_score, fill_value=0.0, dtype=np.float64)
2009
2010
        else:
            return self
Guolin Ke's avatar
Guolin Ke committed
2011
        self.set_init_score(init_score)
2012
        return self
Guolin Ke's avatar
Guolin Ke committed
2013

2014
2015
    def _lazy_init(
        self,
2016
        data: Optional[_LGBM_TrainDataType],
2017
2018
2019
2020
2021
2022
2023
2024
        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,
2025
2026
        params: Optional[Dict[str, Any]],
        position: Optional[_LGBM_PositionType]
2027
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2028
        if data is None:
2029
            self._handle = None
Nikita Titov's avatar
Nikita Titov committed
2030
            return self
Guolin Ke's avatar
Guolin Ke committed
2031
2032
2033
        if reference is not None:
            self.pandas_categorical = reference.pandas_categorical
            categorical_feature = reference.categorical_feature
2034
2035
2036
2037
2038
2039
2040
        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,
                pandas_categorical=self.pandas_categorical
            )
Guolin Ke's avatar
Guolin Ke committed
2041

2042
        # process for args
wxchan's avatar
wxchan committed
2043
        params = {} if params is None else params
2044
        args_names = inspect.signature(self.__class__._lazy_init).parameters.keys()
2045
        for key in params.keys():
2046
            if key in args_names:
2047
2048
                _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.')
2049
        # get categorical features
2050
        if isinstance(categorical_feature, list):
2051
2052
            categorical_indices = set()
            feature_dict = {}
2053
            if isinstance(feature_name, list):
2054
2055
                feature_dict = {name: i for i, name in enumerate(feature_name)}
            for name in categorical_feature:
2056
                if isinstance(name, str) and name in feature_dict:
2057
                    categorical_indices.add(feature_dict[name])
2058
                elif isinstance(name, int):
2059
2060
                    categorical_indices.add(name)
                else:
2061
                    raise TypeError(f"Wrong type({type(name).__name__}) or unknown name({name}) in categorical_feature")
2062
            if categorical_indices:
2063
2064
                for cat_alias in _ConfigAliases.get("categorical_feature"):
                    if cat_alias in params:
2065
                        # If the params[cat_alias] is equal to categorical_indices, do not report the warning.
2066
                        if not (isinstance(params[cat_alias], list) and set(params[cat_alias]) == categorical_indices):
2067
                            _log_warning(f'{cat_alias} in param dict is overridden.')
2068
                        params.pop(cat_alias, None)
2069
                params['categorical_column'] = sorted(categorical_indices)
2070

2071
        params_str = _param_dict_to_str(params)
2072
        self.params = params
2073
        # process for reference dataset
wxchan's avatar
wxchan committed
2074
        ref_dataset = None
wxchan's avatar
wxchan committed
2075
        if isinstance(reference, Dataset):
2076
            ref_dataset = reference.construct()._handle
wxchan's avatar
wxchan committed
2077
2078
        elif reference is not None:
            raise TypeError('Reference dataset should be None or dataset instance')
2079
        # start construct data
2080
        if isinstance(data, (str, Path)):
2081
            self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2082
            _safe_call(_LIB.LGBM_DatasetCreateFromFile(
2083
2084
                _c_str(str(data)),
                _c_str(params_str),
wxchan's avatar
wxchan committed
2085
                ref_dataset,
2086
                ctypes.byref(self._handle)))
wxchan's avatar
wxchan committed
2087
2088
        elif isinstance(data, scipy.sparse.csr_matrix):
            self.__init_from_csr(data, params_str, ref_dataset)
Guolin Ke's avatar
Guolin Ke committed
2089
2090
        elif isinstance(data, scipy.sparse.csc_matrix):
            self.__init_from_csc(data, params_str, ref_dataset)
wxchan's avatar
wxchan committed
2091
2092
        elif isinstance(data, np.ndarray):
            self.__init_from_np2d(data, params_str, ref_dataset)
2093
2094
2095
        elif _is_pyarrow_table(data):
            self.__init_from_pyarrow_table(data, params_str, ref_dataset)
            feature_name = data.column_names
2096
        elif isinstance(data, list) and len(data) > 0:
2097
            if _is_list_of_numpy_arrays(data):
2098
                self.__init_from_list_np2d(data, params_str, ref_dataset)
2099
            elif _is_list_of_sequences(data):
2100
2101
2102
2103
2104
                self.__init_from_seqs(data, ref_dataset)
            else:
                raise TypeError('Data list can only be of ndarray or Sequence')
        elif isinstance(data, Sequence):
            self.__init_from_seqs([data], ref_dataset)
2105
        elif isinstance(data, dt_DataTable):
2106
            self.__init_from_np2d(data.to_numpy(), params_str, ref_dataset)
wxchan's avatar
wxchan committed
2107
2108
2109
2110
        else:
            try:
                csr = scipy.sparse.csr_matrix(data)
                self.__init_from_csr(csr, params_str, ref_dataset)
2111
2112
            except BaseException as err:
                raise TypeError(f'Cannot initialize Dataset from {type(data).__name__}') from err
wxchan's avatar
wxchan committed
2113
2114
2115
        if label is not None:
            self.set_label(label)
        if self.get_label() is None:
2116
            raise ValueError("Label should not be None")
wxchan's avatar
wxchan committed
2117
2118
2119
2120
        if weight is not None:
            self.set_weight(weight)
        if group is not None:
            self.set_group(group)
2121
2122
        if position is not None:
            self.set_position(position)
2123
2124
        if isinstance(predictor, _InnerPredictor):
            if self._predictor is None and init_score is not None:
2125
                _log_warning("The init_score will be overridden by the prediction of init_model.")
2126
2127
2128
2129
2130
            self._set_init_score_by_predictor(
                predictor=predictor,
                data=data,
                used_indices=None
            )
2131
2132
        elif init_score is not None:
            self.set_init_score(init_score)
Guolin Ke's avatar
Guolin Ke committed
2133
        elif predictor is not None:
2134
            raise TypeError(f'Wrong predictor type {type(predictor).__name__}')
Guolin Ke's avatar
Guolin Ke committed
2135
        # set feature names
Nikita Titov's avatar
Nikita Titov committed
2136
        return self.set_feature_name(feature_name)
wxchan's avatar
wxchan committed
2137

2138
2139
    @staticmethod
    def _yield_row_from_seqlist(seqs: List[Sequence], indices: Iterable[int]):
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
        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]
            yield row if row.flags['OWNDATA'] else row.copy()

    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.
2165
        sampled = np.array(list(self._yield_row_from_seqlist(seqs, indices)))
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
        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

2181
2182
2183
    def __init_from_seqs(
        self,
        seqs: List[Sequence],
2184
        ref_dataset: Optional[_DatasetHandle]
2185
    ) -> "Dataset":
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
        """
        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:
2200
            param_str = _param_dict_to_str(self.get_params())
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
            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)
            batch_size = getattr(seq, 'batch_size', None) or Sequence.batch_size
            for start in range(0, nrow, batch_size):
                end = min(start + batch_size, nrow)
                self._push_rows(seq[start:end])
        return self

2214
2215
2216
2217
2218
2219
    def __init_from_np2d(
        self,
        mat: np.ndarray,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
2220
        """Initialize data from a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
2221
2222
2223
        if len(mat.shape) != 2:
            raise ValueError('Input numpy.ndarray must be 2 dimensional')

2224
        self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2225
2226
        if mat.dtype == np.float32 or mat.dtype == np.float64:
            data = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
2227
        else:  # change non-float data to float data, need to copy
wxchan's avatar
wxchan committed
2228
2229
            data = np.array(mat.reshape(mat.size), dtype=np.float32)

2230
        ptr_data, type_ptr_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
2231
2232
        _safe_call(_LIB.LGBM_DatasetCreateFromMat(
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2233
            ctypes.c_int(type_ptr_data),
2234
2235
            ctypes.c_int32(mat.shape[0]),
            ctypes.c_int32(mat.shape[1]),
2236
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
2237
            _c_str(params_str),
wxchan's avatar
wxchan committed
2238
            ref_dataset,
2239
            ctypes.byref(self._handle)))
Nikita Titov's avatar
Nikita Titov committed
2240
        return self
wxchan's avatar
wxchan committed
2241

2242
2243
2244
2245
2246
2247
    def __init_from_list_np2d(
        self,
        mats: List[np.ndarray],
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
2248
        """Initialize data from a list of 2-D numpy matrices."""
2249
        ncol = mats[0].shape[1]
2250
        nrow = np.empty((len(mats),), np.int32)
2251
        ptr_data: _ctypes_float_array
2252
2253
2254
2255
2256
2257
        if mats[0].dtype == np.float64:
            ptr_data = (ctypes.POINTER(ctypes.c_double) * len(mats))()
        else:
            ptr_data = (ctypes.POINTER(ctypes.c_float) * len(mats))()

        holders = []
2258
        type_ptr_data = -1
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270

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

            if mat.shape[1] != ncol:
                raise ValueError('Input arrays must have same number of columns')

            nrow[i] = mat.shape[0]

            if mat.dtype == np.float32 or mat.dtype == np.float64:
                mats[i] = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
2271
            else:  # change non-float data to float data, need to copy
2272
2273
                mats[i] = np.array(mat.reshape(mat.size), dtype=np.float32)

2274
            chunk_ptr_data, chunk_type_ptr_data, holder = _c_float_array(mats[i])
2275
            if type_ptr_data != -1 and chunk_type_ptr_data != type_ptr_data:
2276
2277
2278
2279
2280
                raise ValueError('Input chunks must have same type')
            ptr_data[i] = chunk_ptr_data
            type_ptr_data = chunk_type_ptr_data
            holders.append(holder)

2281
        self._handle = ctypes.c_void_p()
2282
        _safe_call(_LIB.LGBM_DatasetCreateFromMats(
2283
            ctypes.c_int32(len(mats)),
2284
2285
2286
            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)),
2287
            ctypes.c_int32(ncol),
2288
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
2289
            _c_str(params_str),
2290
            ref_dataset,
2291
            ctypes.byref(self._handle)))
Nikita Titov's avatar
Nikita Titov committed
2292
        return self
2293

2294
2295
2296
2297
2298
2299
    def __init_from_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
2300
        """Initialize data from a CSR matrix."""
wxchan's avatar
wxchan committed
2301
        if len(csr.indices) != len(csr.data):
2302
            raise ValueError(f'Length mismatch: {len(csr.indices)} vs {len(csr.data)}')
2303
        self._handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
2304

2305
2306
        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
2307

2308
        assert csr.shape[1] <= _MAX_INT32
2309
        csr_indices = csr.indices.astype(np.int32, copy=False)
2310

wxchan's avatar
wxchan committed
2311
2312
        _safe_call(_LIB.LGBM_DatasetCreateFromCSR(
            ptr_indptr,
Guolin Ke's avatar
Guolin Ke committed
2313
            ctypes.c_int(type_ptr_indptr),
2314
            csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
wxchan's avatar
wxchan committed
2315
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2316
2317
2318
2319
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csr.indptr)),
            ctypes.c_int64(len(csr.data)),
            ctypes.c_int64(csr.shape[1]),
2320
            _c_str(params_str),
wxchan's avatar
wxchan committed
2321
            ref_dataset,
2322
            ctypes.byref(self._handle)))
Nikita Titov's avatar
Nikita Titov committed
2323
        return self
wxchan's avatar
wxchan committed
2324

2325
2326
2327
2328
2329
2330
    def __init_from_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
2331
        """Initialize data from a CSC matrix."""
Guolin Ke's avatar
Guolin Ke committed
2332
        if len(csc.indices) != len(csc.data):
2333
            raise ValueError(f'Length mismatch: {len(csc.indices)} vs {len(csc.data)}')
2334
        self._handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
2335

2336
2337
        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
2338

2339
        assert csc.shape[0] <= _MAX_INT32
2340
        csc_indices = csc.indices.astype(np.int32, copy=False)
2341

Guolin Ke's avatar
Guolin Ke committed
2342
2343
        _safe_call(_LIB.LGBM_DatasetCreateFromCSC(
            ptr_indptr,
Guolin Ke's avatar
Guolin Ke committed
2344
            ctypes.c_int(type_ptr_indptr),
2345
            csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
Guolin Ke's avatar
Guolin Ke committed
2346
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2347
2348
2349
2350
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csc.indptr)),
            ctypes.c_int64(len(csc.data)),
            ctypes.c_int64(csc.shape[0]),
2351
            _c_str(params_str),
Guolin Ke's avatar
Guolin Ke committed
2352
            ref_dataset,
2353
            ctypes.byref(self._handle)))
Nikita Titov's avatar
Nikita Titov committed
2354
        return self
Guolin Ke's avatar
Guolin Ke committed
2355

2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
    def __init_from_pyarrow_table(
        self,
        table: pa_Table,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
        """Initialize data from a PyArrow table."""
        if not PYARROW_INSTALLED:
            raise LightGBMError("Cannot init dataframe from Arrow without `pyarrow` installed.")

        # Check that the input is valid: we only handle numbers (for now)
        if not all(arrow_is_integer(t) or arrow_is_floating(t) for t in table.schema.types):
            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()
        _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)))
        return self

2382
    @staticmethod
2383
    def _compare_params_for_warning(
2384
2385
        params: Dict[str, Any],
        other_params: Dict[str, Any],
2386
2387
2388
        ignore_keys: Set[str]
    ) -> bool:
        """Compare two dictionaries with params ignoring some keys.
2389

2390
2391
2392
2393
        It is only for the warning purpose.

        Parameters
        ----------
2394
        params : dict
2395
            One dictionary with parameters to compare.
2396
        other_params : dict
2397
2398
2399
            Another dictionary with parameters to compare.
        ignore_keys : set
            Keys that should be ignored during comparing two dictionaries.
2400
2401
2402

        Returns
        -------
2403
2404
        compare_result : bool
          Returns whether two dictionaries with params are equal.
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
        """
        for k in other_params:
            if k not in ignore_keys:
                if k not in params or params[k] != other_params[k]:
                    return False
        for k in params:
            if k not in ignore_keys:
                if k not in other_params or params[k] != other_params[k]:
                    return False
        return True

2416
    def construct(self) -> "Dataset":
2417
2418
2419
2420
2421
        """Lazy init.

        Returns
        -------
        self : Dataset
Nikita Titov's avatar
Nikita Titov committed
2422
            Constructed Dataset object.
2423
        """
2424
        if self._handle is None:
wxchan's avatar
wxchan committed
2425
            if self.reference is not None:
2426
                reference_params = self.reference.get_params()
2427
2428
                params = self.get_params()
                if params != reference_params:
2429
2430
2431
2432
2433
                    if not self._compare_params_for_warning(
                        params=params,
                        other_params=reference_params,
                        ignore_keys=_ConfigAliases.get("categorical_feature")
                    ):
2434
                        _log_warning('Overriding the parameters from Reference Dataset.')
2435
                    self._update_params(reference_params)
wxchan's avatar
wxchan committed
2436
                if self.used_indices is None:
2437
                    # create valid
2438
                    self._lazy_init(data=self.data, label=self.label, reference=self.reference,
2439
                                    weight=self.weight, group=self.group, position=self.position,
2440
                                    init_score=self.init_score, predictor=self._predictor,
2441
                                    feature_name=self.feature_name, categorical_feature='auto', params=self.params)
wxchan's avatar
wxchan committed
2442
                else:
2443
                    # construct subset
2444
                    used_indices = _list_to_1d_numpy(self.used_indices, dtype=np.int32, name='used_indices')
2445
                    assert used_indices.flags.c_contiguous
Guolin Ke's avatar
Guolin Ke committed
2446
                    if self.reference.group is not None:
2447
                        group_info = np.array(self.reference.group).astype(np.int32, copy=False)
2448
                        _, self.group = np.unique(np.repeat(range(len(group_info)), repeats=group_info)[self.used_indices],
2449
                                                  return_counts=True)
2450
                    self._handle = ctypes.c_void_p()
2451
                    params_str = _param_dict_to_str(self.params)
wxchan's avatar
wxchan committed
2452
                    _safe_call(_LIB.LGBM_DatasetGetSubset(
2453
                        self.reference.construct()._handle,
wxchan's avatar
wxchan committed
2454
                        used_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
2455
                        ctypes.c_int32(used_indices.shape[0]),
2456
                        _c_str(params_str),
2457
                        ctypes.byref(self._handle)))
Guolin Ke's avatar
Guolin Ke committed
2458
2459
                    if not self.free_raw_data:
                        self.get_data()
Guolin Ke's avatar
Guolin Ke committed
2460
2461
                    if self.group is not None:
                        self.set_group(self.group)
2462
2463
                    if self.position is not None:
                        self.set_position(self.position)
wxchan's avatar
wxchan committed
2464
2465
                    if self.get_label() is None:
                        raise ValueError("Label should not be None.")
Guolin Ke's avatar
Guolin Ke committed
2466
2467
                    if isinstance(self._predictor, _InnerPredictor) and self._predictor is not self.reference._predictor:
                        self.get_data()
2468
2469
2470
2471
2472
                        self._set_init_score_by_predictor(
                            predictor=self._predictor,
                            data=self.data,
                            used_indices=used_indices
                        )
wxchan's avatar
wxchan committed
2473
            else:
2474
                # create train
2475
                self._lazy_init(data=self.data, label=self.label, reference=None,
2476
2477
                                weight=self.weight, group=self.group,
                                init_score=self.init_score, predictor=self._predictor,
2478
2479
                                feature_name=self.feature_name, categorical_feature=self.categorical_feature,
                                params=self.params, position=self.position)
wxchan's avatar
wxchan committed
2480
2481
            if self.free_raw_data:
                self.data = None
2482
            self.feature_name = self.get_feature_name()
wxchan's avatar
wxchan committed
2483
        return self
wxchan's avatar
wxchan committed
2484

2485
2486
    def create_valid(
        self,
2487
        data: _LGBM_TrainDataType,
2488
        label: Optional[_LGBM_LabelType] = None,
2489
2490
2491
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
2492
2493
        params: Optional[Dict[str, Any]] = None,
        position: Optional[_LGBM_PositionType] = None
2494
    ) -> "Dataset":
2495
        """Create validation data align with current Dataset.
wxchan's avatar
wxchan committed
2496
2497
2498

        Parameters
        ----------
2499
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array
wxchan's avatar
wxchan committed
2500
            Data source of Dataset.
2501
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
2502
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2503
            Label of the data.
2504
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2505
            Weight for each instance. Weights should be non-negative.
2506
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
2507
2508
2509
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2510
2511
            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.
2512
        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)
2513
            Init score for Dataset.
Nikita Titov's avatar
Nikita Titov committed
2514
        params : dict or None, optional (default=None)
2515
            Other parameters for validation Dataset.
2516
2517
        position : numpy 1-D array, pandas Series or None, optional (default=None)
            Position of items used in unbiased learning-to-rank task.
2518
2519
2520

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2521
2522
        valid : Dataset
            Validation Dataset with reference to self.
wxchan's avatar
wxchan committed
2523
        """
2524
        ret = Dataset(data, label=label, reference=self,
2525
                      weight=weight, group=group, position=position, init_score=init_score,
2526
                      params=params, free_raw_data=self.free_raw_data)
wxchan's avatar
wxchan committed
2527
        ret._predictor = self._predictor
2528
        ret.pandas_categorical = self.pandas_categorical
wxchan's avatar
wxchan committed
2529
        return ret
wxchan's avatar
wxchan committed
2530

2531
2532
2533
2534
2535
    def subset(
        self,
        used_indices: List[int],
        params: Optional[Dict[str, Any]] = None
    ) -> "Dataset":
2536
        """Get subset of current Dataset.
wxchan's avatar
wxchan committed
2537
2538
2539
2540

        Parameters
        ----------
        used_indices : list of int
2541
            Indices used to create the subset.
Nikita Titov's avatar
Nikita Titov committed
2542
        params : dict or None, optional (default=None)
2543
            These parameters will be passed to Dataset constructor.
2544
2545
2546
2547
2548

        Returns
        -------
        subset : Dataset
            Subset of the current Dataset.
wxchan's avatar
wxchan committed
2549
        """
wxchan's avatar
wxchan committed
2550
2551
        if params is None:
            params = self.params
wxchan's avatar
wxchan committed
2552
        ret = Dataset(None, reference=self, feature_name=self.feature_name,
2553
2554
                      categorical_feature=self.categorical_feature, params=params,
                      free_raw_data=self.free_raw_data)
wxchan's avatar
wxchan committed
2555
        ret._predictor = self._predictor
2556
        ret.pandas_categorical = self.pandas_categorical
2557
        ret.used_indices = sorted(used_indices)
wxchan's avatar
wxchan committed
2558
2559
        return ret

2560
    def save_binary(self, filename: Union[str, Path]) -> "Dataset":
2561
        """Save Dataset to a binary file.
wxchan's avatar
wxchan committed
2562

2563
2564
2565
2566
2567
        .. 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
2568
2569
        Parameters
        ----------
2570
        filename : str or pathlib.Path
wxchan's avatar
wxchan committed
2571
            Name of the output file.
Nikita Titov's avatar
Nikita Titov committed
2572
2573
2574
2575
2576

        Returns
        -------
        self : Dataset
            Returns self.
wxchan's avatar
wxchan committed
2577
2578
        """
        _safe_call(_LIB.LGBM_DatasetSaveBinary(
2579
            self.construct()._handle,
2580
            _c_str(str(filename))))
Nikita Titov's avatar
Nikita Titov committed
2581
        return self
wxchan's avatar
wxchan committed
2582

2583
    def _update_params(self, params: Optional[Dict[str, Any]]) -> "Dataset":
2584
2585
        if not params:
            return self
2586
        params = deepcopy(params)
2587
2588
2589
2590
2591

        def update():
            if not self.params:
                self.params = params
            else:
2592
                self._params_back_up = deepcopy(self.params)
2593
2594
                self.params.update(params)

2595
        if self._handle is None:
2596
2597
2598
            update()
        elif params is not None:
            ret = _LIB.LGBM_DatasetUpdateParamChecking(
2599
2600
                _c_str(_param_dict_to_str(self.params)),
                _c_str(_param_dict_to_str(params)))
2601
2602
2603
2604
2605
2606
            if ret != 0:
                # could be updated if data is not freed
                if self.data is not None:
                    update()
                    self._free_handle()
                else:
2607
                    raise LightGBMError(_LIB.LGBM_GetLastError().decode('utf-8'))
Nikita Titov's avatar
Nikita Titov committed
2608
        return self
wxchan's avatar
wxchan committed
2609

2610
    def _reverse_update_params(self) -> "Dataset":
2611
        if self._handle is None:
2612
2613
            self.params = deepcopy(self._params_back_up)
            self._params_back_up = None
Nikita Titov's avatar
Nikita Titov committed
2614
        return self
2615

2616
2617
2618
    def set_field(
        self,
        field_name: str,
2619
        data: Optional[Union[List[List[float]], List[List[int]], List[float], List[int], np.ndarray, pd_Series, pd_DataFrame, pa_Table, pa_Array, pa_ChunkedArray]]
2620
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2621
        """Set property into the Dataset.
wxchan's avatar
wxchan committed
2622
2623
2624

        Parameters
        ----------
2625
        field_name : str
2626
            The field name of the information.
2627
        data : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), pyarrow Array, pyarrow ChunkedArray or None
2628
            The data to be set.
Nikita Titov's avatar
Nikita Titov committed
2629
2630
2631
2632
2633

        Returns
        -------
        self : Dataset
            Dataset with set property.
wxchan's avatar
wxchan committed
2634
        """
2635
        if self._handle is None:
2636
            raise Exception(f"Cannot set {field_name} before construct dataset")
wxchan's avatar
wxchan committed
2637
        if data is None:
2638
            # set to None
wxchan's avatar
wxchan committed
2639
            _safe_call(_LIB.LGBM_DatasetSetField(
2640
                self._handle,
2641
                _c_str(field_name),
wxchan's avatar
wxchan committed
2642
                None,
Guolin Ke's avatar
Guolin Ke committed
2643
                ctypes.c_int(0),
2644
                ctypes.c_int(_FIELD_TYPE_MAPPER[field_name])))
Nikita Titov's avatar
Nikita Titov committed
2645
            return self
2646
2647

        # If the data is a arrow data, we can just pass it to C
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
        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}'")
                data = pa_chunked_array([
                    chunk for array in data.columns for chunk in array.chunks  # type: ignore
                ])

2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
            c_array = _export_arrow_to_c(data)
            _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),
            ))
            self.version += 1
            return self

2669
        dtype: "np.typing.DTypeLike"
2670
        if field_name == 'init_score':
Guolin Ke's avatar
Guolin Ke committed
2671
            dtype = np.float64
2672
            if _is_1d_collection(data):
2673
                data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2674
            elif _is_2d_collection(data):
2675
                data = _data_to_2d_numpy(data, dtype=dtype, name=field_name)
2676
2677
2678
2679
2680
2681
2682
                data = data.ravel(order='F')
            else:
                raise TypeError(
                    '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.'
                )
        else:
2683
2684
2685
2686
            if field_name in {'group', 'position'}:
                dtype = np.int32
            else:
                dtype = np.float32
2687
            data = _list_to_1d_numpy(data, dtype=dtype, name=field_name)
2688

2689
        ptr_data: Union[_ctypes_float_ptr, _ctypes_int_ptr]
2690
        if data.dtype == np.float32 or data.dtype == np.float64:
2691
            ptr_data, type_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
2692
        elif data.dtype == np.int32:
2693
            ptr_data, type_data, _ = _c_int_array(data)
wxchan's avatar
wxchan committed
2694
        else:
2695
            raise TypeError(f"Expected np.float32/64 or np.int32, met type({data.dtype})")
2696
        if type_data != _FIELD_TYPE_MAPPER[field_name]:
2697
            raise TypeError("Input type error for set_field")
wxchan's avatar
wxchan committed
2698
        _safe_call(_LIB.LGBM_DatasetSetField(
2699
            self._handle,
2700
            _c_str(field_name),
wxchan's avatar
wxchan committed
2701
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2702
2703
            ctypes.c_int(len(data)),
            ctypes.c_int(type_data)))
2704
        self.version += 1
Nikita Titov's avatar
Nikita Titov committed
2705
        return self
wxchan's avatar
wxchan committed
2706

2707
    def get_field(self, field_name: str) -> Optional[np.ndarray]:
wxchan's avatar
wxchan committed
2708
        """Get property from the Dataset.
wxchan's avatar
wxchan committed
2709

2710
2711
2712
2713
2714
2715
        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
2716
2717
        Parameters
        ----------
2718
        field_name : str
2719
            The field name of the information.
wxchan's avatar
wxchan committed
2720
2721
2722

        Returns
        -------
2723
        info : numpy array or None
2724
            A numpy array with information from the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2725
        """
2726
        if self._handle is None:
2727
            raise Exception(f"Cannot get {field_name} before construct Dataset")
2728
2729
        tmp_out_len = ctypes.c_int(0)
        out_type = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2730
2731
        ret = ctypes.POINTER(ctypes.c_void_p)()
        _safe_call(_LIB.LGBM_DatasetGetField(
2732
            self._handle,
2733
            _c_str(field_name),
wxchan's avatar
wxchan committed
2734
2735
2736
            ctypes.byref(tmp_out_len),
            ctypes.byref(ret),
            ctypes.byref(out_type)))
2737
        if out_type.value != _FIELD_TYPE_MAPPER[field_name]:
wxchan's avatar
wxchan committed
2738
2739
2740
            raise TypeError("Return type error for get_field")
        if tmp_out_len.value == 0:
            return None
2741
        if out_type.value == _C_API_DTYPE_INT32:
2742
2743
2744
2745
            arr = _cint32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_int32)),
                length=tmp_out_len.value
            )
2746
        elif out_type.value == _C_API_DTYPE_FLOAT32:
2747
2748
2749
2750
            arr = _cfloat32_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_float)),
                length=tmp_out_len.value
            )
2751
        elif out_type.value == _C_API_DTYPE_FLOAT64:
2752
2753
2754
2755
            arr = _cfloat64_array_to_numpy(
                cptr=ctypes.cast(ret, ctypes.POINTER(ctypes.c_double)),
                length=tmp_out_len.value
            )
2756
        else:
wxchan's avatar
wxchan committed
2757
            raise TypeError("Unknown type")
2758
2759
2760
2761
2762
2763
        if field_name == 'init_score':
            num_data = self.num_data()
            num_classes = arr.size // num_data
            if num_classes > 1:
                arr = arr.reshape((num_data, num_classes), order='F')
        return arr
Guolin Ke's avatar
Guolin Ke committed
2764

2765
2766
    def set_categorical_feature(
        self,
2767
        categorical_feature: _LGBM_CategoricalFeatureConfiguration
2768
    ) -> "Dataset":
2769
        """Set categorical features.
2770
2771
2772

        Parameters
        ----------
2773
        categorical_feature : list of str or int, or 'auto'
2774
            Names or indices of categorical features.
Nikita Titov's avatar
Nikita Titov committed
2775
2776
2777
2778
2779

        Returns
        -------
        self : Dataset
            Dataset with set categorical features.
2780
2781
        """
        if self.categorical_feature == categorical_feature:
Nikita Titov's avatar
Nikita Titov committed
2782
            return self
2783
        if self.data is not None:
2784
2785
            if self.categorical_feature is None:
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2786
                return self._free_handle()
2787
            elif categorical_feature == 'auto':
Nikita Titov's avatar
Nikita Titov committed
2788
                return self
2789
            else:
2790
2791
                if self.categorical_feature != 'auto':
                    _log_warning('categorical_feature in Dataset is overridden.\n'
2792
                                 f'New categorical_feature is {list(categorical_feature)}')
2793
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2794
                return self._free_handle()
2795
        else:
2796
2797
            raise LightGBMError("Cannot set categorical feature after freed raw data, "
                                "set free_raw_data=False when construct Dataset to avoid this.")
2798

2799
2800
2801
2802
    def _set_predictor(
        self,
        predictor: Optional[_InnerPredictor]
    ) -> "Dataset":
2803
2804
2805
2806
        """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
2807
        """
2808
        if predictor is None and self._predictor is None:
Nikita Titov's avatar
Nikita Titov committed
2809
            return self
2810
2811
2812
        elif isinstance(predictor, _InnerPredictor) and isinstance(self._predictor, _InnerPredictor):
            if (predictor == self._predictor) and (predictor.current_iteration() == self._predictor.current_iteration()):
                return self
2813
        if self._handle is None:
Guolin Ke's avatar
Guolin Ke committed
2814
            self._predictor = predictor
2815
2816
        elif self.data is not None:
            self._predictor = predictor
2817
2818
2819
2820
2821
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.data,
                used_indices=None
            )
2822
2823
        elif self.used_indices is not None and self.reference is not None and self.reference.data is not None:
            self._predictor = predictor
2824
2825
2826
2827
2828
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.reference.data,
                used_indices=self.used_indices
            )
Guolin Ke's avatar
Guolin Ke committed
2829
        else:
2830
2831
            raise LightGBMError("Cannot set predictor after freed raw data, "
                                "set free_raw_data=False when construct Dataset to avoid this.")
2832
        return self
Guolin Ke's avatar
Guolin Ke committed
2833

2834
    def set_reference(self, reference: "Dataset") -> "Dataset":
2835
        """Set reference Dataset.
Guolin Ke's avatar
Guolin Ke committed
2836
2837
2838
2839

        Parameters
        ----------
        reference : Dataset
2840
            Reference that is used as a template to construct the current Dataset.
Nikita Titov's avatar
Nikita Titov committed
2841
2842
2843
2844
2845

        Returns
        -------
        self : Dataset
            Dataset with set reference.
Guolin Ke's avatar
Guolin Ke committed
2846
        """
2847
2848
2849
        self.set_categorical_feature(reference.categorical_feature) \
            .set_feature_name(reference.feature_name) \
            ._set_predictor(reference._predictor)
2850
        # we're done if self and reference share a common upstream reference
2851
        if self.get_ref_chain().intersection(reference.get_ref_chain()):
Nikita Titov's avatar
Nikita Titov committed
2852
            return self
Guolin Ke's avatar
Guolin Ke committed
2853
2854
        if self.data is not None:
            self.reference = reference
Nikita Titov's avatar
Nikita Titov committed
2855
            return self._free_handle()
Guolin Ke's avatar
Guolin Ke committed
2856
        else:
2857
2858
            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
2859

2860
    def set_feature_name(self, feature_name: _LGBM_FeatureNameConfiguration) -> "Dataset":
2861
        """Set feature name.
Guolin Ke's avatar
Guolin Ke committed
2862
2863
2864

        Parameters
        ----------
2865
        feature_name : list of str
2866
            Feature names.
Nikita Titov's avatar
Nikita Titov committed
2867
2868
2869
2870
2871

        Returns
        -------
        self : Dataset
            Dataset with set feature name.
Guolin Ke's avatar
Guolin Ke committed
2872
        """
2873
2874
        if feature_name != 'auto':
            self.feature_name = feature_name
2875
        if self._handle is not None and feature_name is not None and feature_name != 'auto':
wxchan's avatar
wxchan committed
2876
            if len(feature_name) != self.num_feature():
2877
                raise ValueError(f"Length of feature_name({len(feature_name)}) and num_feature({self.num_feature()}) don't match")
2878
            c_feature_name = [_c_str(name) for name in feature_name]
wxchan's avatar
wxchan committed
2879
            _safe_call(_LIB.LGBM_DatasetSetFeatureNames(
2880
                self._handle,
2881
                _c_array(ctypes.c_char_p, c_feature_name),
Guolin Ke's avatar
Guolin Ke committed
2882
                ctypes.c_int(len(feature_name))))
Nikita Titov's avatar
Nikita Titov committed
2883
        return self
Guolin Ke's avatar
Guolin Ke committed
2884

2885
    def set_label(self, label: Optional[_LGBM_LabelType]) -> "Dataset":
2886
        """Set label of Dataset.
Guolin Ke's avatar
Guolin Ke committed
2887
2888
2889

        Parameters
        ----------
2890
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array, pyarrow ChunkedArray or None
2891
            The label information to be set into Dataset.
Nikita Titov's avatar
Nikita Titov committed
2892
2893
2894
2895
2896

        Returns
        -------
        self : Dataset
            Dataset with set label.
Guolin Ke's avatar
Guolin Ke committed
2897
2898
        """
        self.label = label
2899
        if self._handle is not None:
2900
2901
2902
            if isinstance(label, pd_DataFrame):
                if len(label.columns) > 1:
                    raise ValueError('DataFrame for label cannot have multiple columns')
2903
                label_array = np.ravel(_pandas_to_numpy(label, target_dtype=np.float32))
2904
2905
            elif _is_pyarrow_array(label):
                label_array = label
2906
            else:
2907
                label_array = _list_to_1d_numpy(label, dtype=np.float32, name='label')
2908
            self.set_field('label', label_array)
2909
            self.label = self.get_field('label')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2910
        return self
Guolin Ke's avatar
Guolin Ke committed
2911

2912
2913
2914
2915
    def set_weight(
        self,
        weight: Optional[_LGBM_WeightType]
    ) -> "Dataset":
2916
        """Set weight of each instance.
Guolin Ke's avatar
Guolin Ke committed
2917
2918
2919

        Parameters
        ----------
2920
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
2921
            Weight to be set for each data point. Weights should be non-negative.
Nikita Titov's avatar
Nikita Titov committed
2922
2923
2924
2925
2926

        Returns
        -------
        self : Dataset
            Dataset with set weight.
Guolin Ke's avatar
Guolin Ke committed
2927
        """
2928
2929
2930
2931
2932
2933
2934
        # 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
2935
        self.weight = weight
2936
2937

        # Set field
2938
        if self._handle is not None and weight is not None:
2939
2940
            if not _is_pyarrow_array(weight):
                weight = _list_to_1d_numpy(weight, dtype=np.float32, name='weight')
wxchan's avatar
wxchan committed
2941
            self.set_field('weight', weight)
2942
            self.weight = self.get_field('weight')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2943
        return self
Guolin Ke's avatar
Guolin Ke committed
2944

2945
2946
2947
2948
    def set_init_score(
        self,
        init_score: Optional[_LGBM_InitScoreType]
    ) -> "Dataset":
2949
        """Set init score of Booster to start from.
Guolin Ke's avatar
Guolin Ke committed
2950
2951
2952

        Parameters
        ----------
2953
        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
2954
            Init score for Booster.
Nikita Titov's avatar
Nikita Titov committed
2955
2956
2957
2958
2959

        Returns
        -------
        self : Dataset
            Dataset with set init score.
Guolin Ke's avatar
Guolin Ke committed
2960
2961
        """
        self.init_score = init_score
2962
        if self._handle is not None and init_score is not None:
wxchan's avatar
wxchan committed
2963
            self.set_field('init_score', init_score)
2964
            self.init_score = self.get_field('init_score')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2965
        return self
Guolin Ke's avatar
Guolin Ke committed
2966

2967
2968
2969
2970
    def set_group(
        self,
        group: Optional[_LGBM_GroupType]
    ) -> "Dataset":
2971
        """Set group size of Dataset (used for ranking).
Guolin Ke's avatar
Guolin Ke committed
2972
2973
2974

        Parameters
        ----------
2975
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None
2976
2977
2978
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2979
2980
            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
2981
2982
2983
2984
2985

        Returns
        -------
        self : Dataset
            Dataset with set group.
Guolin Ke's avatar
Guolin Ke committed
2986
2987
        """
        self.group = group
2988
        if self._handle is not None and group is not None:
2989
2990
            if not _is_pyarrow_array(group):
                group = _list_to_1d_numpy(group, dtype=np.int32, name='group')
wxchan's avatar
wxchan committed
2991
            self.set_field('group', group)
2992
2993
2994
2995
            # original values can be modified at cpp side
            constructed_group = self.get_field('group')
            if constructed_group is not None:
                self.group = np.diff(constructed_group)
Nikita Titov's avatar
Nikita Titov committed
2996
        return self
Guolin Ke's avatar
Guolin Ke committed
2997

2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
    def set_position(
        self,
        position: Optional[_LGBM_PositionType]
    ) -> "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:
            position = _list_to_1d_numpy(position, dtype=np.int32, name='position')
            self.set_field('position', position)
        return self

3020
    def get_feature_name(self) -> List[str]:
3021
3022
3023
3024
        """Get the names of columns (features) in the Dataset.

        Returns
        -------
3025
        feature_names : list of str
3026
3027
            The names of columns (features) in the Dataset.
        """
3028
        if self._handle is None:
3029
3030
3031
3032
3033
            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)
3034
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
3035
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3036
        _safe_call(_LIB.LGBM_DatasetGetFeatureNames(
3037
            self._handle,
3038
            ctypes.c_int(num_feature),
3039
            ctypes.byref(tmp_out_len),
3040
            ctypes.c_size_t(reserved_string_buffer_size),
3041
3042
3043
3044
            ctypes.byref(required_string_buffer_size),
            ptr_string_buffers))
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
3045
3046
3047
3048
        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)]
3049
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
3050
            _safe_call(_LIB.LGBM_DatasetGetFeatureNames(
3051
                self._handle,
3052
3053
3054
3055
3056
                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))
3057
        return [string_buffers[i].value.decode('utf-8') for i in range(num_feature)]
3058

3059
    def get_label(self) -> Optional[_LGBM_LabelType]:
3060
        """Get the label of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3061
3062
3063

        Returns
        -------
3064
        label : list, numpy 1-D array, pandas Series / one-column DataFrame or None
3065
            The label information from the Dataset.
3066
            For a constructed ``Dataset``, this will only return a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3067
        """
3068
        if self.label is None:
wxchan's avatar
wxchan committed
3069
            self.label = self.get_field('label')
Guolin Ke's avatar
Guolin Ke committed
3070
3071
        return self.label

3072
    def get_weight(self) -> Optional[_LGBM_WeightType]:
3073
        """Get the weight of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3074
3075
3076

        Returns
        -------
3077
        weight : list, numpy 1-D array, pandas Series or None
3078
            Weight for each data point from the Dataset. Weights should be non-negative.
3079
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3080
        """
3081
        if self.weight is None:
wxchan's avatar
wxchan committed
3082
            self.weight = self.get_field('weight')
Guolin Ke's avatar
Guolin Ke committed
3083
3084
        return self.weight

3085
    def get_init_score(self) -> Optional[_LGBM_InitScoreType]:
3086
        """Get the initial score of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3087
3088
3089

        Returns
        -------
3090
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None
3091
            Init score of Booster.
3092
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3093
        """
3094
        if self.init_score is None:
wxchan's avatar
wxchan committed
3095
            self.init_score = self.get_field('init_score')
Guolin Ke's avatar
Guolin Ke committed
3096
3097
        return self.init_score

3098
    def get_data(self) -> Optional[_LGBM_TrainDataType]:
3099
3100
3101
3102
        """Get the raw data of the Dataset.

        Returns
        -------
3103
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array or None
3104
3105
            Raw data used in the Dataset construction.
        """
3106
        if self._handle is None:
3107
            raise Exception("Cannot get data before construct Dataset")
3108
        if self._need_slice and self.used_indices is not None and self.reference is not None:
Guolin Ke's avatar
Guolin Ke committed
3109
3110
            self.data = self.reference.data
            if self.data is not None:
3111
                if isinstance(self.data, (np.ndarray, scipy.sparse.spmatrix)):
Guolin Ke's avatar
Guolin Ke committed
3112
                    self.data = self.data[self.used_indices, :]
3113
                elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3114
                    self.data = self.data.iloc[self.used_indices].copy()
3115
                elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3116
                    self.data = self.data[self.used_indices, :]
3117
3118
                elif isinstance(self.data, Sequence):
                    self.data = self.data[self.used_indices]
3119
                elif _is_list_of_sequences(self.data) and len(self.data) > 0:
3120
                    self.data = np.array(list(self._yield_row_from_seqlist(self.data, self.used_indices)))
Guolin Ke's avatar
Guolin Ke committed
3121
                else:
3122
3123
                    _log_warning(f"Cannot subset {type(self.data).__name__} type of raw data.\n"
                                 "Returning original raw data")
3124
            self._need_slice = False
Guolin Ke's avatar
Guolin Ke committed
3125
3126
3127
        if self.data is None:
            raise LightGBMError("Cannot call `get_data` after freed raw data, "
                                "set free_raw_data=False when construct Dataset to avoid this.")
3128
3129
        return self.data

3130
    def get_group(self) -> Optional[_LGBM_GroupType]:
3131
        """Get the group of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3132
3133
3134

        Returns
        -------
3135
        group : list, numpy 1-D array, pandas Series or None
3136
3137
3138
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
3139
3140
            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.
3141
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
Guolin Ke's avatar
Guolin Ke committed
3142
        """
3143
        if self.group is None:
wxchan's avatar
wxchan committed
3144
            self.group = self.get_field('group')
Guolin Ke's avatar
Guolin Ke committed
3145
3146
            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
3147
                self.group = np.diff(self.group)
Guolin Ke's avatar
Guolin Ke committed
3148
3149
        return self.group

3150
    def get_position(self) -> Optional[_LGBM_PositionType]:
3151
3152
3153
3154
        """Get the position of the Dataset.

        Returns
        -------
3155
        position : numpy 1-D array, pandas Series or None
3156
            Position of items used in unbiased learning-to-rank task.
3157
            For a constructed ``Dataset``, this will only return ``None`` or a numpy array.
3158
3159
3160
3161
3162
        """
        if self.position is None:
            self.position = self.get_field('position')
        return self.position

3163
    def num_data(self) -> int:
3164
        """Get the number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3165
3166
3167

        Returns
        -------
3168
3169
        number_of_rows : int
            The number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3170
        """
3171
        if self._handle is not None:
3172
            ret = ctypes.c_int(0)
3173
            _safe_call(_LIB.LGBM_DatasetGetNumData(self._handle,
wxchan's avatar
wxchan committed
3174
3175
                                                   ctypes.byref(ret)))
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3176
        else:
3177
            raise LightGBMError("Cannot get num_data before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3178

3179
    def num_feature(self) -> int:
3180
        """Get the number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3181
3182
3183

        Returns
        -------
3184
3185
        number_of_columns : int
            The number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
3186
        """
3187
        if self._handle is not None:
3188
            ret = ctypes.c_int(0)
3189
            _safe_call(_LIB.LGBM_DatasetGetNumFeature(self._handle,
wxchan's avatar
wxchan committed
3190
3191
                                                      ctypes.byref(ret)))
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
3192
        else:
3193
            raise LightGBMError("Cannot get num_feature before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
3194

3195
    def feature_num_bin(self, feature: Union[int, str]) -> int:
3196
3197
        """Get the number of bins for a feature.

3198
3199
        .. versionadded:: 4.0.0

3200
3201
        Parameters
        ----------
3202
3203
        feature : int or str
            Index or name of the feature.
3204
3205
3206
3207
3208
3209

        Returns
        -------
        number_of_bins : int
            The number of constructed bins for the feature in the Dataset.
        """
3210
        if self._handle is not None:
3211
            if isinstance(feature, str):
3212
3213
3214
                feature_index = self.feature_name.index(feature)
            else:
                feature_index = feature
3215
            ret = ctypes.c_int(0)
3216
            _safe_call(_LIB.LGBM_DatasetGetFeatureNumBin(self._handle,
3217
                                                         ctypes.c_int(feature_index),
3218
3219
3220
3221
3222
                                                         ctypes.byref(ret)))
            return ret.value
        else:
            raise LightGBMError("Cannot get feature_num_bin before construct dataset")

3223
    def get_ref_chain(self, ref_limit: int = 100) -> Set["Dataset"]:
3224
3225
3226
3227
3228
        """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.
3229
3230
3231
3232
3233

        Parameters
        ----------
        ref_limit : int, optional (default=100)
            The limit number of references.
3234
3235
3236

        Returns
        -------
3237
3238
3239
        ref_chain : set of Dataset
            Chain of references of the Datasets.
        """
3240
        head = self
3241
        ref_chain: Set[Dataset] = set()
3242
3243
        while len(ref_chain) < ref_limit:
            if isinstance(head, Dataset):
3244
                ref_chain.add(head)
3245
3246
3247
3248
3249
3250
                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
3251
        return ref_chain
3252

3253
    def add_features_from(self, other: "Dataset") -> "Dataset":
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
        """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.
        """
3268
        if self._handle is None or other._handle is None:
3269
            raise ValueError('Both source and target Datasets must be constructed before adding features')
3270
        _safe_call(_LIB.LGBM_DatasetAddFeaturesFrom(self._handle, other._handle))
Guolin Ke's avatar
Guolin Ke committed
3271
3272
3273
3274
3275
3276
3277
3278
        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))
3279
                elif isinstance(other.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3280
                    self.data = np.hstack((self.data, other.data.toarray()))
3281
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3282
                    self.data = np.hstack((self.data, other.data.values))
3283
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3284
3285
3286
                    self.data = np.hstack((self.data, other.data.to_numpy()))
                else:
                    self.data = None
3287
            elif isinstance(self.data, scipy.sparse.spmatrix):
Guolin Ke's avatar
Guolin Ke committed
3288
                sparse_format = self.data.getformat()
3289
                if isinstance(other.data, (np.ndarray, scipy.sparse.spmatrix)):
Guolin Ke's avatar
Guolin Ke committed
3290
                    self.data = scipy.sparse.hstack((self.data, other.data), format=sparse_format)
3291
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3292
                    self.data = scipy.sparse.hstack((self.data, other.data.values), format=sparse_format)
3293
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3294
3295
3296
                    self.data = scipy.sparse.hstack((self.data, other.data.to_numpy()), format=sparse_format)
                else:
                    self.data = None
3297
            elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3298
3299
                if not PANDAS_INSTALLED:
                    raise LightGBMError("Cannot add features to DataFrame type of raw data "
3300
3301
                                        "without pandas installed. "
                                        "Install pandas and restart your session.")
Guolin Ke's avatar
Guolin Ke committed
3302
                if isinstance(other.data, np.ndarray):
3303
                    self.data = concat((self.data, pd_DataFrame(other.data)),
Guolin Ke's avatar
Guolin Ke committed
3304
                                       axis=1, ignore_index=True)
3305
                elif isinstance(other.data, scipy.sparse.spmatrix):
3306
                    self.data = concat((self.data, pd_DataFrame(other.data.toarray())),
Guolin Ke's avatar
Guolin Ke committed
3307
                                       axis=1, ignore_index=True)
3308
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
3309
3310
                    self.data = concat((self.data, other.data),
                                       axis=1, ignore_index=True)
3311
3312
                elif isinstance(other.data, dt_DataTable):
                    self.data = concat((self.data, pd_DataFrame(other.data.to_numpy())),
Guolin Ke's avatar
Guolin Ke committed
3313
3314
3315
                                       axis=1, ignore_index=True)
                else:
                    self.data = None
3316
            elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
3317
                if isinstance(other.data, np.ndarray):
3318
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data)))
3319
                elif isinstance(other.data, scipy.sparse.spmatrix):
3320
3321
3322
3323
3324
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.toarray())))
                elif isinstance(other.data, pd_DataFrame):
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.values)))
                elif isinstance(other.data, dt_DataTable):
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data.to_numpy())))
Guolin Ke's avatar
Guolin Ke committed
3325
3326
3327
3328
3329
                else:
                    self.data = None
            else:
                self.data = None
        if self.data is None:
3330
3331
            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")
Guolin Ke's avatar
Guolin Ke committed
3332
3333
            err_msg += ("Set free_raw_data=False when construct Dataset to avoid this"
                        if was_none else "Freeing raw data")
3334
            _log_warning(err_msg)
Guolin Ke's avatar
Guolin Ke committed
3335
        self.feature_name = self.get_feature_name()
3336
3337
        _log_warning("Reseting categorical features.\n"
                     "You can set new categorical features via ``set_categorical_feature`` method")
Guolin Ke's avatar
Guolin Ke committed
3338
3339
        self.categorical_feature = "auto"
        self.pandas_categorical = None
3340
3341
        return self

3342
    def _dump_text(self, filename: Union[str, Path]) -> "Dataset":
3343
3344
3345
3346
3347
3348
        """Save Dataset to a text file.

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

        Parameters
        ----------
3349
        filename : str or pathlib.Path
3350
3351
3352
3353
3354
3355
3356
3357
            Name of the output file.

        Returns
        -------
        self : Dataset
            Returns self.
        """
        _safe_call(_LIB.LGBM_DatasetDumpText(
3358
            self.construct()._handle,
3359
            _c_str(str(filename))))
3360
3361
        return self

wxchan's avatar
wxchan committed
3362

3363
3364
3365
3366
_LGBM_CustomObjectiveFunction = Callable[
    [np.ndarray, Dataset],
    Tuple[np.ndarray, np.ndarray]
]
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
_LGBM_CustomEvalFunction = Union[
    Callable[
        [np.ndarray, Dataset],
        _LGBM_EvalFunctionResultType
    ],
    Callable[
        [np.ndarray, Dataset],
        List[_LGBM_EvalFunctionResultType]
    ]
]
3377
3378


3379
class Booster:
3380
    """Booster in LightGBM."""
3381

3382
3383
3384
3385
3386
3387
3388
    def __init__(
        self,
        params: Optional[Dict[str, Any]] = None,
        train_set: Optional[Dataset] = None,
        model_file: Optional[Union[str, Path]] = None,
        model_str: Optional[str] = None
    ):
3389
        """Initialize the Booster.
wxchan's avatar
wxchan committed
3390
3391
3392

        Parameters
        ----------
Nikita Titov's avatar
Nikita Titov committed
3393
        params : dict or None, optional (default=None)
3394
3395
3396
            Parameters for Booster.
        train_set : Dataset or None, optional (default=None)
            Training dataset.
3397
        model_file : str, pathlib.Path or None, optional (default=None)
wxchan's avatar
wxchan committed
3398
            Path to the model file.
3399
        model_str : str or None, optional (default=None)
3400
            Model will be loaded from this string.
wxchan's avatar
wxchan committed
3401
        """
3402
        self._handle = ctypes.c_void_p()
3403
        self._network = False
wxchan's avatar
wxchan committed
3404
        self.__need_reload_eval_info = True
3405
        self._train_data_name = "training"
3406
        self.__set_objective_to_none = False
wxchan's avatar
wxchan committed
3407
        self.best_iteration = -1
3408
        self.best_score: _LGBM_BoosterBestScoreType = {}
3409
        params = {} if params is None else deepcopy(params)
wxchan's avatar
wxchan committed
3410
        if train_set is not None:
3411
            # Training task
wxchan's avatar
wxchan committed
3412
            if not isinstance(train_set, Dataset):
3413
                raise TypeError(f'Training data should be Dataset instance, met {type(train_set).__name__}')
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
            params = _choose_param_value(
                main_param_name="machines",
                params=params,
                default_value=None
            )
            # 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):
                    num_machines_from_machine_list = len(machines.split(','))
                elif isinstance(machines, (list, set)):
                    num_machines_from_machine_list = len(machines)
                    machines = ','.join(machines)
                else:
                    raise ValueError("Invalid machines in params.")

                params = _choose_param_value(
                    main_param_name="num_machines",
                    params=params,
                    default_value=num_machines_from_machine_list
                )
                params = _choose_param_value(
                    main_param_name="local_listen_port",
                    params=params,
                    default_value=12400
                )
                self.set_network(
                    machines=machines,
                    local_listen_port=params["local_listen_port"],
                    listen_time_out=params.get("time_out", 120),
                    num_machines=params["num_machines"]
                )
3448
            # construct booster object
3449
3450
3451
            train_set.construct()
            # copy the parameters from train_set
            params.update(train_set.get_params())
3452
            params_str = _param_dict_to_str(params)
wxchan's avatar
wxchan committed
3453
            _safe_call(_LIB.LGBM_BoosterCreate(
3454
                train_set._handle,
3455
                _c_str(params_str),
3456
                ctypes.byref(self._handle)))
3457
            # save reference to data
wxchan's avatar
wxchan committed
3458
            self.train_set = train_set
3459
3460
            self.valid_sets: List[Dataset] = []
            self.name_valid_sets: List[str] = []
wxchan's avatar
wxchan committed
3461
            self.__num_dataset = 1
Guolin Ke's avatar
Guolin Ke committed
3462
3463
            self.__init_predictor = train_set._predictor
            if self.__init_predictor is not None:
wxchan's avatar
wxchan committed
3464
                _safe_call(_LIB.LGBM_BoosterMerge(
3465
3466
                    self._handle,
                    self.__init_predictor._handle))
Guolin Ke's avatar
Guolin Ke committed
3467
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3468
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
3469
                self._handle,
wxchan's avatar
wxchan committed
3470
3471
                ctypes.byref(out_num_class)))
            self.__num_class = out_num_class.value
3472
            # buffer for inner predict
3473
            self.__inner_predict_buffer: List[Optional[np.ndarray]] = [None]
wxchan's avatar
wxchan committed
3474
3475
            self.__is_predicted_cur_iter = [False]
            self.__get_eval_info()
3476
            self.pandas_categorical = train_set.pandas_categorical
3477
            self.train_set_version = train_set.version
wxchan's avatar
wxchan committed
3478
        elif model_file is not None:
3479
            # Prediction task
Guolin Ke's avatar
Guolin Ke committed
3480
            out_num_iterations = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3481
            _safe_call(_LIB.LGBM_BoosterCreateFromModelfile(
3482
                _c_str(str(model_file)),
wxchan's avatar
wxchan committed
3483
                ctypes.byref(out_num_iterations),
3484
                ctypes.byref(self._handle)))
Guolin Ke's avatar
Guolin Ke committed
3485
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3486
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
3487
                self._handle,
wxchan's avatar
wxchan committed
3488
3489
                ctypes.byref(out_num_class)))
            self.__num_class = out_num_class.value
3490
            self.pandas_categorical = _load_pandas_categorical(file_name=model_file)
3491
3492
3493
            if params:
                _log_warning('Ignoring params argument, using parameters from model file.')
            params = self._get_loaded_param()
3494
        elif model_str is not None:
3495
            self.model_from_string(model_str)
wxchan's avatar
wxchan committed
3496
        else:
3497
3498
            raise TypeError('Need at least one training dataset or model file or model string '
                            'to create Booster instance')
3499
        self.params = params
wxchan's avatar
wxchan committed
3500

3501
    def __del__(self) -> None:
3502
        try:
3503
            if self._network:
3504
3505
3506
3507
                self.free_network()
        except AttributeError:
            pass
        try:
3508
3509
            if self._handle is not None:
                _safe_call(_LIB.LGBM_BoosterFree(self._handle))
3510
3511
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
3512

3513
    def __copy__(self) -> "Booster":
wxchan's avatar
wxchan committed
3514
3515
        return self.__deepcopy__(None)

3516
    def __deepcopy__(self, _) -> "Booster":
3517
        model_str = self.model_to_string(num_iteration=-1)
3518
        return Booster(model_str=model_str)
wxchan's avatar
wxchan committed
3519

3520
    def __getstate__(self) -> Dict[str, Any]:
wxchan's avatar
wxchan committed
3521
        this = self.__dict__.copy()
3522
        handle = this['_handle']
wxchan's avatar
wxchan committed
3523
3524
3525
        this.pop('train_set', None)
        this.pop('valid_sets', None)
        if handle is not None:
3526
            this["_handle"] = self.model_to_string(num_iteration=-1)
wxchan's avatar
wxchan committed
3527
3528
        return this

3529
    def __setstate__(self, state: Dict[str, Any]) -> None:
3530
        model_str = state.get('_handle', state.get('handle', None))
3531
        if model_str is not None:
wxchan's avatar
wxchan committed
3532
            handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
3533
            out_num_iterations = ctypes.c_int(0)
3534
            _safe_call(_LIB.LGBM_BoosterLoadModelFromString(
3535
                _c_str(model_str),
3536
3537
                ctypes.byref(out_num_iterations),
                ctypes.byref(handle)))
3538
            state['_handle'] = handle
wxchan's avatar
wxchan committed
3539
3540
        self.__dict__.update(state)

3541
3542
3543
3544
    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)
3545
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3546
        _safe_call(_LIB.LGBM_BoosterGetLoadedParam(
3547
            self._handle,
3548
3549
3550
3551
3552
3553
3554
            ctypes.c_int64(buffer_len),
            ctypes.byref(tmp_out_len),
            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)
3555
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
3556
            _safe_call(_LIB.LGBM_BoosterGetLoadedParam(
3557
                self._handle,
3558
3559
3560
3561
3562
                ctypes.c_int64(actual_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
        return json.loads(string_buffer.value.decode('utf-8'))

3563
    def free_dataset(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3564
3565
3566
3567
3568
3569
3570
        """Free Booster's Datasets.

        Returns
        -------
        self : Booster
            Booster without Datasets.
        """
wxchan's avatar
wxchan committed
3571
3572
        self.__dict__.pop('train_set', None)
        self.__dict__.pop('valid_sets', None)
3573
        self.__num_dataset = 0
Nikita Titov's avatar
Nikita Titov committed
3574
        return self
wxchan's avatar
wxchan committed
3575

3576
    def _free_buffer(self) -> "Booster":
3577
3578
        self.__inner_predict_buffer = []
        self.__is_predicted_cur_iter = []
Nikita Titov's avatar
Nikita Titov committed
3579
        return self
3580

3581
3582
3583
3584
3585
3586
3587
    def set_network(
        self,
        machines: Union[List[str], Set[str], str],
        local_listen_port: int = 12400,
        listen_time_out: int = 120,
        num_machines: int = 1
    ) -> "Booster":
3588
3589
3590
3591
        """Set the network configuration.

        Parameters
        ----------
3592
        machines : list, set or str
3593
            Names of machines.
Nikita Titov's avatar
Nikita Titov committed
3594
        local_listen_port : int, optional (default=12400)
3595
            TCP listen port for local machines.
Nikita Titov's avatar
Nikita Titov committed
3596
        listen_time_out : int, optional (default=120)
3597
            Socket time-out in minutes.
Nikita Titov's avatar
Nikita Titov committed
3598
        num_machines : int, optional (default=1)
3599
            The number of machines for distributed learning application.
Nikita Titov's avatar
Nikita Titov committed
3600
3601
3602
3603
3604

        Returns
        -------
        self : Booster
            Booster with set network.
3605
        """
3606
3607
        if isinstance(machines, (list, set)):
            machines = ','.join(machines)
3608
        _safe_call(_LIB.LGBM_NetworkInit(_c_str(machines),
3609
3610
3611
                                         ctypes.c_int(local_listen_port),
                                         ctypes.c_int(listen_time_out),
                                         ctypes.c_int(num_machines)))
3612
        self._network = True
Nikita Titov's avatar
Nikita Titov committed
3613
        return self
3614

3615
    def free_network(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3616
3617
3618
3619
3620
3621
3622
        """Free Booster's network.

        Returns
        -------
        self : Booster
            Booster with freed network.
        """
3623
        _safe_call(_LIB.LGBM_NetworkFree())
3624
        self._network = False
Nikita Titov's avatar
Nikita Titov committed
3625
        return self
3626

3627
    def trees_to_dataframe(self) -> pd_DataFrame:
3628
3629
        """Parse the fitted model and return in an easy-to-read pandas DataFrame.

3630
3631
3632
3633
        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.
3634
3635
3636
3637
3638
            - ``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.
3639
3640
            - ``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.
3641
            - ``decision_type`` : str, logical operator describing how to compare a value to ``threshold``.
3642
3643
              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.
3644
3645
            - ``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.
3646
            - ``value`` : float64, predicted value for this leaf node, multiplied by the learning rate.
3647
            - ``weight`` : float64 or int64, sum of Hessian (second-order derivative of objective), summed over observations that fall in this node.
3648
3649
            - ``count`` : int64, number of records in the training data that fall into this node.

3650
3651
3652
3653
3654
3655
        Returns
        -------
        result : pandas DataFrame
            Returns a pandas DataFrame of the parsed model.
        """
        if not PANDAS_INSTALLED:
3656
3657
            raise LightGBMError('This method cannot be run without pandas installed. '
                                'You must install pandas and restart your session to use this method.')
3658
3659
3660
3661

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

3662
        def _is_split_node(tree: Dict[str, Any]) -> bool:
3663
3664
            return 'split_index' in tree.keys()

3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
        def create_node_record(
            tree: Dict[str, Any],
            node_depth: int = 1,
            tree_index: Optional[int] = None,
            feature_names: Optional[List[str]] = None,
            parent_node: Optional[str] = None
        ) -> Dict[str, Any]:

            def _get_node_index(
                tree: Dict[str, Any],
                tree_index: Optional[int]
            ) -> str:
3677
                tree_num = f'{tree_index}-' if tree_index is not None else ''
3678
3679
3680
                is_split = _is_split_node(tree)
                node_type = 'S' if is_split else 'L'
                # if a single node tree it won't have `leaf_index` so return 0
3681
3682
                node_num = tree.get('split_index' if is_split else 'leaf_index', 0)
                return f"{tree_num}{node_type}{node_num}"
3683

3684
3685
3686
3687
            def _get_split_feature(
                tree: Dict[str, Any],
                feature_names: Optional[List[str]]
            ) -> Optional[str]:
3688
3689
3690
3691
3692
3693
3694
3695
3696
                if _is_split_node(tree):
                    if feature_names is not None:
                        feature_name = feature_names[tree['split_feature']]
                    else:
                        feature_name = tree['split_feature']
                else:
                    feature_name = None
                return feature_name

3697
            def _is_single_node_tree(tree: Dict[str, Any]) -> bool:
3698
                return set(tree.keys()) == {'leaf_value'}
3699
3700

            # Create the node record, and populate universal data members
3701
            node: Dict[str, Union[int, str, None]] = OrderedDict()
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
            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

            # Update values to reflect node type (leaf or split)
            if _is_split_node(tree):
                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']
            else:
                node['value'] = tree['leaf_value']
                if not _is_single_node_tree(tree):
                    node['weight'] = tree['leaf_weight']
                    node['count'] = tree['leaf_count']

            return node

3738
3739
3740
3741
3742
3743
3744
        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,
            parent_node: Optional[str] = None
        ) -> List[Dict[str, Any]]:
3745

3746
            node = create_node_record(tree=tree,
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
                                      node_depth=node_depth,
                                      tree_index=tree_index,
                                      feature_names=feature_names,
                                      parent_node=parent_node)

            res = [node]

            if _is_split_node(tree):
                # traverse the next level of the tree
                children = ['left_child', 'right_child']
                for child in children:
                    subtree_list = tree_dict_to_node_list(
3759
                        tree=tree[child],
3760
3761
3762
                        node_depth=node_depth + 1,
                        tree_index=tree_index,
                        feature_names=feature_names,
3763
3764
                        parent_node=node['node_index']
                    )
3765
3766
3767
3768
3769
3770
3771
3772
3773
                    # 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()
        feature_names = model_dict['feature_names']
        model_list = []
        for tree in model_dict['tree_info']:
3774
            model_list.extend(tree_dict_to_node_list(tree=tree['tree_structure'],
3775
3776
3777
                                                     tree_index=tree['tree_index'],
                                                     feature_names=feature_names))

3778
        return pd_DataFrame(model_list, columns=model_list[0].keys())
3779

3780
    def set_train_data_name(self, name: str) -> "Booster":
3781
3782
3783
3784
        """Set the name to the training Dataset.

        Parameters
        ----------
3785
        name : str
Nikita Titov's avatar
Nikita Titov committed
3786
3787
3788
3789
3790
3791
            Name for the training Dataset.

        Returns
        -------
        self : Booster
            Booster with set training Dataset name.
3792
        """
3793
        self._train_data_name = name
Nikita Titov's avatar
Nikita Titov committed
3794
        return self
wxchan's avatar
wxchan committed
3795

3796
    def add_valid(self, data: Dataset, name: str) -> "Booster":
3797
        """Add validation data.
wxchan's avatar
wxchan committed
3798
3799
3800
3801

        Parameters
        ----------
        data : Dataset
3802
            Validation data.
3803
        name : str
3804
            Name of validation data.
Nikita Titov's avatar
Nikita Titov committed
3805
3806
3807
3808
3809

        Returns
        -------
        self : Booster
            Booster with set validation data.
wxchan's avatar
wxchan committed
3810
        """
Guolin Ke's avatar
Guolin Ke committed
3811
        if not isinstance(data, Dataset):
3812
            raise TypeError(f'Validation data should be Dataset instance, met {type(data).__name__}')
Guolin Ke's avatar
Guolin Ke committed
3813
        if data._predictor is not self.__init_predictor:
3814
3815
            raise LightGBMError("Add validation data failed, "
                                "you should use same predictor for these data")
wxchan's avatar
wxchan committed
3816
        _safe_call(_LIB.LGBM_BoosterAddValidData(
3817
3818
            self._handle,
            data.construct()._handle))
wxchan's avatar
wxchan committed
3819
3820
3821
3822
3823
        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
3824
        return self
wxchan's avatar
wxchan committed
3825

3826
    def reset_parameter(self, params: Dict[str, Any]) -> "Booster":
3827
        """Reset parameters of Booster.
wxchan's avatar
wxchan committed
3828
3829
3830
3831

        Parameters
        ----------
        params : dict
3832
            New parameters for Booster.
Nikita Titov's avatar
Nikita Titov committed
3833
3834
3835
3836
3837

        Returns
        -------
        self : Booster
            Booster with new parameters.
wxchan's avatar
wxchan committed
3838
        """
3839
        params_str = _param_dict_to_str(params)
wxchan's avatar
wxchan committed
3840
3841
        if params_str:
            _safe_call(_LIB.LGBM_BoosterResetParameter(
3842
                self._handle,
3843
                _c_str(params_str)))
Guolin Ke's avatar
Guolin Ke committed
3844
        self.params.update(params)
Nikita Titov's avatar
Nikita Titov committed
3845
        return self
wxchan's avatar
wxchan committed
3846

3847
3848
3849
3850
3851
    def update(
        self,
        train_set: Optional[Dataset] = None,
        fobj: Optional[_LGBM_CustomObjectiveFunction] = None
    ) -> bool:
Nikita Titov's avatar
Nikita Titov committed
3852
        """Update Booster for one iteration.
3853

wxchan's avatar
wxchan committed
3854
3855
        Parameters
        ----------
3856
3857
3858
3859
        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
3860
            Customized objective function.
3861
3862
3863
            Should accept two parameters: preds, train_data,
            and return (grad, hess).

3864
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
3865
                    The predicted values.
3866
3867
                    Predicted values are returned before any transformation,
                    e.g. they are raw margin instead of probability of positive class for binary task.
3868
3869
                train_data : Dataset
                    The training dataset.
3870
                grad : numpy 1-D array or numpy 2-D array (for multi-class task)
3871
3872
                    The value of the first order derivative (gradient) of the loss
                    with respect to the elements of preds for each sample point.
3873
                hess : numpy 1-D array or numpy 2-D array (for multi-class task)
3874
3875
                    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
3876

3877
            For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes],
3878
            and grad and hess should be returned in the same format.
3879

wxchan's avatar
wxchan committed
3880
3881
        Returns
        -------
3882
3883
        is_finished : bool
            Whether the update was successfully finished.
wxchan's avatar
wxchan committed
3884
        """
3885
        # need reset training data
3886
3887
3888
3889
3890
3891
        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
3892
            if not isinstance(train_set, Dataset):
3893
                raise TypeError(f'Training data should be Dataset instance, met {type(train_set).__name__}')
Guolin Ke's avatar
Guolin Ke committed
3894
            if train_set._predictor is not self.__init_predictor:
3895
3896
                raise LightGBMError("Replace training data failed, "
                                    "you should use same predictor for these data")
wxchan's avatar
wxchan committed
3897
3898
            self.train_set = train_set
            _safe_call(_LIB.LGBM_BoosterResetTrainingData(
3899
3900
                self._handle,
                self.train_set.construct()._handle))
wxchan's avatar
wxchan committed
3901
            self.__inner_predict_buffer[0] = None
3902
            self.train_set_version = self.train_set.version
wxchan's avatar
wxchan committed
3903
3904
        is_finished = ctypes.c_int(0)
        if fobj is None:
3905
            if self.__set_objective_to_none:
3906
                raise LightGBMError('Cannot update due to null objective function.')
wxchan's avatar
wxchan committed
3907
            _safe_call(_LIB.LGBM_BoosterUpdateOneIter(
3908
                self._handle,
wxchan's avatar
wxchan committed
3909
                ctypes.byref(is_finished)))
3910
            self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
3911
3912
            return is_finished.value == 1
        else:
3913
            if not self.__set_objective_to_none:
Nikita Titov's avatar
Nikita Titov committed
3914
                self.reset_parameter({"objective": "none"}).__set_objective_to_none = True
wxchan's avatar
wxchan committed
3915
3916
3917
            grad, hess = fobj(self.__inner_predict(0), self.train_set)
            return self.__boost(grad, hess)

3918
3919
3920
3921
3922
    def __boost(
        self,
        grad: np.ndarray,
        hess: np.ndarray
    ) -> bool:
3923
        """Boost Booster for one iteration with customized gradient statistics.
Nikita Titov's avatar
Nikita Titov committed
3924

Nikita Titov's avatar
Nikita Titov committed
3925
3926
        .. note::

3927
3928
            Score is returned before any transformation,
            e.g. it is raw margin instead of probability of positive class for binary task.
3929
            For multi-class task, score are numpy 2-D array of shape = [n_samples, n_classes],
3930
            and grad and hess should be returned in the same format.
3931

wxchan's avatar
wxchan committed
3932
3933
        Parameters
        ----------
3934
        grad : numpy 1-D array or numpy 2-D array (for multi-class task)
3935
3936
            The value of the first order derivative (gradient) of the loss
            with respect to the elements of score for each sample point.
3937
        hess : numpy 1-D array or numpy 2-D array (for multi-class task)
3938
3939
            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
3940
3941
3942

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3943
3944
        is_finished : bool
            Whether the boost was successfully finished.
wxchan's avatar
wxchan committed
3945
        """
3946
3947
3948
        if self.__num_class > 1:
            grad = grad.ravel(order='F')
            hess = hess.ravel(order='F')
3949
3950
        grad = _list_to_1d_numpy(grad, dtype=np.float32, name='gradient')
        hess = _list_to_1d_numpy(hess, dtype=np.float32, name='hessian')
3951
3952
        assert grad.flags.c_contiguous
        assert hess.flags.c_contiguous
wxchan's avatar
wxchan committed
3953
        if len(grad) != len(hess):
3954
3955
            raise ValueError(f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) don't match")
        num_train_data = self.train_set.num_data()
3956
        if len(grad) != num_train_data * self.__num_class:
3957
3958
3959
            raise ValueError(
                f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) "
                f"don't match training data length ({num_train_data}) * "
3960
                f"number of models per one iteration ({self.__num_class})"
3961
            )
wxchan's avatar
wxchan committed
3962
3963
        is_finished = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterUpdateOneIterCustom(
3964
            self._handle,
wxchan's avatar
wxchan committed
3965
3966
3967
            grad.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
            hess.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
            ctypes.byref(is_finished)))
3968
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
3969
3970
        return is_finished.value == 1

3971
    def rollback_one_iter(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3972
3973
3974
3975
3976
3977
3978
        """Rollback one iteration.

        Returns
        -------
        self : Booster
            Booster with rolled back one iteration.
        """
wxchan's avatar
wxchan committed
3979
        _safe_call(_LIB.LGBM_BoosterRollbackOneIter(
3980
            self._handle))
3981
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
Nikita Titov's avatar
Nikita Titov committed
3982
        return self
wxchan's avatar
wxchan committed
3983

3984
    def current_iteration(self) -> int:
3985
3986
3987
3988
3989
3990
3991
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
Guolin Ke's avatar
Guolin Ke committed
3992
        out_cur_iter = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3993
        _safe_call(_LIB.LGBM_BoosterGetCurrentIteration(
3994
            self._handle,
wxchan's avatar
wxchan committed
3995
3996
3997
            ctypes.byref(out_cur_iter)))
        return out_cur_iter.value

3998
    def num_model_per_iteration(self) -> int:
3999
4000
4001
4002
4003
4004
4005
4006
4007
        """Get number of models per iteration.

        Returns
        -------
        model_per_iter : int
            The number of models per iteration.
        """
        model_per_iter = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterNumModelPerIteration(
4008
            self._handle,
4009
4010
4011
            ctypes.byref(model_per_iter)))
        return model_per_iter.value

4012
    def num_trees(self) -> int:
4013
4014
4015
4016
4017
4018
4019
4020
4021
        """Get number of weak sub-models.

        Returns
        -------
        num_trees : int
            The number of weak sub-models.
        """
        num_trees = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterNumberOfTotalModel(
4022
            self._handle,
4023
4024
4025
            ctypes.byref(num_trees)))
        return num_trees.value

4026
    def upper_bound(self) -> float:
4027
4028
4029
4030
        """Get upper bound value of a model.

        Returns
        -------
4031
        upper_bound : float
4032
4033
4034
4035
            Upper bound value of the model.
        """
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetUpperBoundValue(
4036
            self._handle,
4037
4038
4039
            ctypes.byref(ret)))
        return ret.value

4040
    def lower_bound(self) -> float:
4041
4042
4043
4044
        """Get lower bound value of a model.

        Returns
        -------
4045
        lower_bound : float
4046
4047
4048
4049
            Lower bound value of the model.
        """
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetLowerBoundValue(
4050
            self._handle,
4051
4052
4053
            ctypes.byref(ret)))
        return ret.value

4054
4055
4056
4057
4058
4059
    def eval(
        self,
        data: Dataset,
        name: str,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4060
        """Evaluate for data.
wxchan's avatar
wxchan committed
4061
4062
4063

        Parameters
        ----------
4064
4065
        data : Dataset
            Data for the evaluating.
4066
        name : str
4067
            Name of the data.
4068
        feval : callable, list of callable, or None, optional (default=None)
4069
            Customized evaluation function.
4070
            Each evaluation function should accept two parameters: preds, eval_data,
4071
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4072

4073
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4074
                    The predicted values.
4075
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4076
                    If custom objective function is used, predicted values are returned before any transformation,
4077
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
4078
                eval_data : Dataset
4079
                    A ``Dataset`` to evaluate.
4080
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4081
                    The name of evaluation function (without whitespace).
4082
4083
4084
4085
4086
                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
4087
4088
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4089
        result : list
4090
            List with (dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4091
        """
Guolin Ke's avatar
Guolin Ke committed
4092
4093
        if not isinstance(data, Dataset):
            raise TypeError("Can only eval for Dataset instance")
wxchan's avatar
wxchan committed
4094
4095
4096
4097
        data_idx = -1
        if data is self.train_set:
            data_idx = 0
        else:
4098
            for i in range(len(self.valid_sets)):
wxchan's avatar
wxchan committed
4099
4100
4101
                if data is self.valid_sets[i]:
                    data_idx = i + 1
                    break
4102
        # need to push new valid data
wxchan's avatar
wxchan committed
4103
4104
4105
4106
4107
4108
        if data_idx == -1:
            self.add_valid(data, name)
            data_idx = self.__num_dataset - 1

        return self.__inner_eval(name, data_idx, feval)

4109
4110
4111
4112
    def eval_train(
        self,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4113
        """Evaluate for training data.
wxchan's avatar
wxchan committed
4114
4115
4116

        Parameters
        ----------
4117
        feval : callable, list of callable, or None, optional (default=None)
4118
            Customized evaluation function.
4119
            Each evaluation function should accept two parameters: preds, eval_data,
4120
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4121

4122
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4123
                    The predicted values.
4124
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4125
                    If custom objective function is used, predicted values are returned before any transformation,
4126
                    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
4127
                eval_data : Dataset
4128
                    The training dataset.
4129
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4130
                    The name of evaluation function (without whitespace).
4131
4132
4133
4134
4135
                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
4136
4137
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4138
        result : list
4139
            List with (train_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4140
        """
4141
        return self.__inner_eval(self._train_data_name, 0, feval)
wxchan's avatar
wxchan committed
4142

4143
4144
4145
4146
    def eval_valid(
        self,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4147
        """Evaluate for validation data.
wxchan's avatar
wxchan committed
4148
4149
4150

        Parameters
        ----------
4151
        feval : callable, list of callable, or None, optional (default=None)
4152
            Customized evaluation function.
4153
            Each evaluation function should accept two parameters: preds, eval_data,
4154
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
4155

4156
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
4157
                    The predicted values.
4158
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
4159
                    If custom objective function is used, predicted values are returned before any transformation,
4160
                    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
4161
                eval_data : Dataset
4162
                    The validation dataset.
4163
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
4164
                    The name of evaluation function (without whitespace).
4165
4166
4167
4168
4169
                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
4170
4171
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4172
        result : list
4173
            List with (validation_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
4174
        """
4175
        return [item for i in range(1, self.__num_dataset)
wxchan's avatar
wxchan committed
4176
                for item in self.__inner_eval(self.name_valid_sets[i - 1], i, feval)]
wxchan's avatar
wxchan committed
4177

4178
4179
4180
4181
4182
4183
4184
    def save_model(
        self,
        filename: Union[str, Path],
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
        importance_type: str = 'split'
    ) -> "Booster":
4185
        """Save Booster to file.
wxchan's avatar
wxchan committed
4186
4187
4188

        Parameters
        ----------
4189
        filename : str or pathlib.Path
4190
            Filename to save Booster.
4191
4192
4193
4194
        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
4195
        start_iteration : int, optional (default=0)
4196
            Start index of the iteration that should be saved.
4197
        importance_type : str, optional (default="split")
4198
4199
4200
            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
4201
4202
4203
4204
4205

        Returns
        -------
        self : Booster
            Returns self.
wxchan's avatar
wxchan committed
4206
        """
4207
        if num_iteration is None:
4208
            num_iteration = self.best_iteration
4209
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
4210
        _safe_call(_LIB.LGBM_BoosterSaveModel(
4211
            self._handle,
4212
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
4213
            ctypes.c_int(num_iteration),
4214
            ctypes.c_int(importance_type_int),
4215
            _c_str(str(filename))))
4216
        _dump_pandas_categorical(self.pandas_categorical, filename)
Nikita Titov's avatar
Nikita Titov committed
4217
        return self
wxchan's avatar
wxchan committed
4218

4219
4220
4221
4222
4223
    def shuffle_models(
        self,
        start_iteration: int = 0,
        end_iteration: int = -1
    ) -> "Booster":
4224
        """Shuffle models.
Nikita Titov's avatar
Nikita Titov committed
4225

4226
4227
4228
        Parameters
        ----------
        start_iteration : int, optional (default=0)
4229
            The first iteration that will be shuffled.
4230
4231
        end_iteration : int, optional (default=-1)
            The last iteration that will be shuffled.
4232
            If <= 0, means the last available iteration.
4233

Nikita Titov's avatar
Nikita Titov committed
4234
4235
4236
4237
        Returns
        -------
        self : Booster
            Booster with shuffled models.
4238
        """
4239
        _safe_call(_LIB.LGBM_BoosterShuffleModels(
4240
            self._handle,
Guolin Ke's avatar
Guolin Ke committed
4241
4242
            ctypes.c_int(start_iteration),
            ctypes.c_int(end_iteration)))
Nikita Titov's avatar
Nikita Titov committed
4243
        return self
4244

4245
    def model_from_string(self, model_str: str) -> "Booster":
4246
4247
4248
4249
        """Load Booster from a string.

        Parameters
        ----------
4250
        model_str : str
4251
4252
4253
4254
            Model will be loaded from this string.

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
4255
        self : Booster
4256
4257
            Loaded Booster object.
        """
4258
4259
4260
        # ensure that existing Booster is freed before replacing it
        # with a new one createdfrom file
        _safe_call(_LIB.LGBM_BoosterFree(self._handle))
4261
        self._free_buffer()
4262
        self._handle = ctypes.c_void_p()
4263
4264
        out_num_iterations = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterLoadModelFromString(
4265
            _c_str(model_str),
4266
            ctypes.byref(out_num_iterations),
4267
            ctypes.byref(self._handle)))
4268
4269
        out_num_class = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterGetNumClasses(
4270
            self._handle,
4271
4272
            ctypes.byref(out_num_class)))
        self.__num_class = out_num_class.value
4273
        self.pandas_categorical = _load_pandas_categorical(model_str=model_str)
4274
4275
        return self

4276
4277
4278
4279
4280
4281
    def model_to_string(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
        importance_type: str = 'split'
    ) -> str:
4282
        """Save Booster to string.
4283

4284
4285
4286
4287
4288
4289
        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
4290
        start_iteration : int, optional (default=0)
4291
            Start index of the iteration that should be saved.
4292
        importance_type : str, optional (default="split")
4293
4294
4295
            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.
4296
4297
4298

        Returns
        -------
4299
        str_repr : str
4300
4301
            String representation of Booster.
        """
4302
        if num_iteration is None:
4303
            num_iteration = self.best_iteration
4304
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4305
        buffer_len = 1 << 20
4306
        tmp_out_len = ctypes.c_int64(0)
4307
        string_buffer = ctypes.create_string_buffer(buffer_len)
4308
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4309
        _safe_call(_LIB.LGBM_BoosterSaveModelToString(
4310
            self._handle,
4311
            ctypes.c_int(start_iteration),
4312
            ctypes.c_int(num_iteration),
4313
            ctypes.c_int(importance_type_int),
4314
            ctypes.c_int64(buffer_len),
4315
4316
4317
            ctypes.byref(tmp_out_len),
            ptr_string_buffer))
        actual_len = tmp_out_len.value
4318
        # if buffer length is not long enough, re-allocate a buffer
4319
4320
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
4321
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
4322
            _safe_call(_LIB.LGBM_BoosterSaveModelToString(
4323
                self._handle,
4324
                ctypes.c_int(start_iteration),
4325
                ctypes.c_int(num_iteration),
4326
                ctypes.c_int(importance_type_int),
4327
                ctypes.c_int64(actual_len),
4328
4329
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
4330
        ret = string_buffer.value.decode('utf-8')
4331
4332
        ret += _dump_pandas_categorical(self.pandas_categorical)
        return ret
4333

4334
4335
4336
4337
4338
4339
4340
    def dump_model(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
        importance_type: str = 'split',
        object_hook: Optional[Callable[[Dict[str, Any]], Dict[str, Any]]] = None
    ) -> Dict[str, Any]:
Nikita Titov's avatar
Nikita Titov committed
4341
        """Dump Booster to JSON format.
wxchan's avatar
wxchan committed
4342

4343
4344
        Parameters
        ----------
4345
4346
4347
4348
        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
4349
        start_iteration : int, optional (default=0)
4350
            Start index of the iteration that should be dumped.
4351
        importance_type : str, optional (default="split")
4352
4353
4354
            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.
4355
4356
4357
4358
4359
4360
4361
4362
4363
        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.
4364

wxchan's avatar
wxchan committed
4365
4366
        Returns
        -------
4367
        json_repr : dict
Nikita Titov's avatar
Nikita Titov committed
4368
            JSON format of Booster.
wxchan's avatar
wxchan committed
4369
        """
4370
        if num_iteration is None:
4371
            num_iteration = self.best_iteration
4372
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
4373
        buffer_len = 1 << 20
4374
        tmp_out_len = ctypes.c_int64(0)
wxchan's avatar
wxchan committed
4375
        string_buffer = ctypes.create_string_buffer(buffer_len)
4376
        ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
wxchan's avatar
wxchan committed
4377
        _safe_call(_LIB.LGBM_BoosterDumpModel(
4378
            self._handle,
4379
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
4380
            ctypes.c_int(num_iteration),
4381
            ctypes.c_int(importance_type_int),
4382
            ctypes.c_int64(buffer_len),
wxchan's avatar
wxchan committed
4383
            ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
4384
            ptr_string_buffer))
wxchan's avatar
wxchan committed
4385
        actual_len = tmp_out_len.value
4386
        # if buffer length is not long enough, reallocate a buffer
wxchan's avatar
wxchan committed
4387
4388
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
4389
            ptr_string_buffer = ctypes.c_char_p(ctypes.addressof(string_buffer))
wxchan's avatar
wxchan committed
4390
            _safe_call(_LIB.LGBM_BoosterDumpModel(
4391
                self._handle,
4392
                ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
4393
                ctypes.c_int(num_iteration),
4394
                ctypes.c_int(importance_type_int),
4395
                ctypes.c_int64(actual_len),
wxchan's avatar
wxchan committed
4396
                ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
4397
                ptr_string_buffer))
4398
        ret = json.loads(string_buffer.value.decode('utf-8'), object_hook=object_hook)
4399
        ret['pandas_categorical'] = json.loads(json.dumps(self.pandas_categorical,
4400
                                                          default=_json_default_with_numpy))
4401
        return ret
wxchan's avatar
wxchan committed
4402

4403
4404
    def predict(
        self,
4405
        data: _LGBM_PredictDataType,
4406
4407
4408
4409
4410
4411
4412
4413
        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,
        **kwargs: Any
4414
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
4415
        """Make a prediction.
wxchan's avatar
wxchan committed
4416
4417
4418

        Parameters
        ----------
4419
        data : str, pathlib.Path, numpy array, pandas DataFrame, pyarrow Table, H2O DataTable's Frame or scipy.sparse
4420
            Data source for prediction.
4421
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4422
        start_iteration : int, optional (default=0)
4423
            Start index of the iteration to predict.
4424
            If <= 0, starts from the first iteration.
4425
        num_iteration : int or None, optional (default=None)
4426
4427
4428
4429
            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).
4430
4431
4432
4433
        raw_score : bool, optional (default=False)
            Whether to predict raw scores.
        pred_leaf : bool, optional (default=False)
            Whether to predict leaf index.
4434
4435
        pred_contrib : bool, optional (default=False)
            Whether to predict feature contributions.
4436

Nikita Titov's avatar
Nikita Titov committed
4437
4438
4439
4440
4441
4442
4443
            .. 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.
4444

4445
4446
        data_has_header : bool, optional (default=False)
            Whether the data has header.
4447
            Used only if data is str.
4448
4449
4450
        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.
4451
4452
        **kwargs
            Other parameters for the prediction.
wxchan's avatar
wxchan committed
4453
4454
4455

        Returns
        -------
4456
        result : numpy array, scipy.sparse or list of scipy.sparse
4457
            Prediction result.
4458
            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
4459
        """
4460
4461
4462
4463
        predictor = _InnerPredictor.from_booster(
            booster=self,
            pred_parameter=deepcopy(kwargs),
        )
4464
        if num_iteration is None:
4465
            if start_iteration <= 0:
4466
4467
4468
                num_iteration = self.best_iteration
            else:
                num_iteration = -1
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
        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,
            validate_features=validate_features
        )
wxchan's avatar
wxchan committed
4479

4480
4481
    def refit(
        self,
4482
        data: _LGBM_TrainDataType,
4483
        label: _LGBM_LabelType,
4484
4485
        decay_rate: float = 0.9,
        reference: Optional[Dataset] = None,
4486
4487
4488
        weight: Optional[_LGBM_WeightType] = None,
        group: Optional[_LGBM_GroupType] = None,
        init_score: Optional[_LGBM_InitScoreType] = None,
4489
4490
        feature_name: _LGBM_FeatureNameConfiguration = 'auto',
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = 'auto',
4491
4492
4493
        dataset_params: Optional[Dict[str, Any]] = None,
        free_raw_data: bool = True,
        validate_features: bool = False,
4494
        **kwargs
4495
    ) -> "Booster":
Guolin Ke's avatar
Guolin Ke committed
4496
4497
4498
4499
        """Refit the existing Booster by new data.

        Parameters
        ----------
4500
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame, scipy.sparse, Sequence, list of Sequence or list of numpy array
Guolin Ke's avatar
Guolin Ke committed
4501
            Data source for refit.
4502
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4503
        label : list, numpy 1-D array, pandas Series / one-column DataFrame, pyarrow Array or pyarrow ChunkedArray
Guolin Ke's avatar
Guolin Ke committed
4504
4505
            Label for refit.
        decay_rate : float, optional (default=0.9)
4506
4507
            Decay rate of refit,
            will use ``leaf_output = decay_rate * old_leaf_output + (1.0 - decay_rate) * new_leaf_output`` to refit trees.
4508
4509
        reference : Dataset or None, optional (default=None)
            Reference for ``data``.
4510
4511
4512

            .. versionadded:: 4.0.0

4513
        weight : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4514
            Weight for each ``data`` instance. Weights should be non-negative.
4515
4516
4517

            .. versionadded:: 4.0.0

4518
        group : list, numpy 1-D array, pandas Series, pyarrow Array, pyarrow ChunkedArray or None, optional (default=None)
4519
4520
4521
4522
4523
            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.
4524
4525
4526

            .. versionadded:: 4.0.0

4527
        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)
4528
            Init score for ``data``.
4529
4530
4531

            .. versionadded:: 4.0.0

4532
4533
4534
        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.
4535
4536
4537

            .. versionadded:: 4.0.0

4538
4539
4540
4541
4542
        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.
4543
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
4544
4545
4546
            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.
4547
            Floating point numbers in categorical features will be rounded towards 0.
4548
4549
4550

            .. versionadded:: 4.0.0

4551
4552
        dataset_params : dict or None, optional (default=None)
            Other parameters for Dataset ``data``.
4553
4554
4555

            .. versionadded:: 4.0.0

4556
4557
        free_raw_data : bool, optional (default=True)
            If True, raw data is freed after constructing inner Dataset for ``data``.
4558
4559
4560

            .. versionadded:: 4.0.0

4561
4562
4563
        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.
4564
4565
4566

            .. versionadded:: 4.0.0

4567
4568
        **kwargs
            Other parameters for refit.
4569
            These parameters will be passed to ``predict`` method.
Guolin Ke's avatar
Guolin Ke committed
4570
4571
4572
4573
4574
4575

        Returns
        -------
        result : Booster
            Refitted Booster.
        """
4576
4577
        if self.__set_objective_to_none:
            raise LightGBMError('Cannot refit due to null objective function.')
4578
4579
        if dataset_params is None:
            dataset_params = {}
4580
4581
4582
4583
        predictor = _InnerPredictor.from_booster(
            booster=self,
            pred_parameter=deepcopy(kwargs)
        )
4584
        leaf_preds: np.ndarray = predictor.predict(  # type: ignore[assignment]
4585
4586
4587
4588
4589
            data=data,
            start_iteration=-1,
            pred_leaf=True,
            validate_features=validate_features
        )
4590
        nrow, ncol = leaf_preds.shape
4591
        out_is_linear = ctypes.c_int(0)
4592
        _safe_call(_LIB.LGBM_BoosterGetLinear(
4593
            self._handle,
4594
            ctypes.byref(out_is_linear)))
Nikita Titov's avatar
Nikita Titov committed
4595
4596
4597
4598
4599
        new_params = _choose_param_value(
            main_param_name="linear_tree",
            params=self.params,
            default_value=None
        )
4600
        new_params["linear_tree"] = bool(out_is_linear.value)
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
        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,
        )
4614
        new_params['refit_decay_rate'] = decay_rate
4615
        new_booster = Booster(new_params, train_set)
Guolin Ke's avatar
Guolin Ke committed
4616
4617
        # Copy models
        _safe_call(_LIB.LGBM_BoosterMerge(
4618
4619
            new_booster._handle,
            predictor._handle))
Guolin Ke's avatar
Guolin Ke committed
4620
        leaf_preds = leaf_preds.reshape(-1)
4621
        ptr_data, _, _ = _c_int_array(leaf_preds)
Guolin Ke's avatar
Guolin Ke committed
4622
        _safe_call(_LIB.LGBM_BoosterRefit(
4623
            new_booster._handle,
Guolin Ke's avatar
Guolin Ke committed
4624
            ptr_data,
4625
4626
            ctypes.c_int32(nrow),
            ctypes.c_int32(ncol)))
4627
        new_booster._network = self._network
Guolin Ke's avatar
Guolin Ke committed
4628
4629
        return new_booster

4630
    def get_leaf_output(self, tree_id: int, leaf_id: int) -> float:
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
        """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.
        """
4645
4646
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetLeafValue(
4647
            self._handle,
4648
4649
4650
4651
4652
            ctypes.c_int(tree_id),
            ctypes.c_int(leaf_id),
            ctypes.byref(ret)))
        return ret.value

4653
4654
4655
4656
4657
4658
4659
4660
    def set_leaf_output(
        self,
        tree_id: int,
        leaf_id: int,
        value: float,
    ) -> 'Booster':
        """Set the output of a leaf.

4661
4662
        .. versionadded:: 4.0.0

4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
        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(
4679
                self._handle,
4680
4681
4682
4683
4684
4685
4686
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
                ctypes.c_double(value)
            )
        )
        return self

4687
    def num_feature(self) -> int:
4688
4689
4690
4691
4692
4693
4694
        """Get number of features.

        Returns
        -------
        num_feature : int
            The number of features.
        """
4695
4696
        out_num_feature = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterGetNumFeature(
4697
            self._handle,
4698
4699
4700
            ctypes.byref(out_num_feature)))
        return out_num_feature.value

4701
    def feature_name(self) -> List[str]:
4702
        """Get names of features.
wxchan's avatar
wxchan committed
4703
4704
4705

        Returns
        -------
4706
        result : list of str
4707
            List with names of features.
wxchan's avatar
wxchan committed
4708
        """
4709
        num_feature = self.num_feature()
4710
        # Get name of features
wxchan's avatar
wxchan committed
4711
        tmp_out_len = ctypes.c_int(0)
4712
4713
        reserved_string_buffer_size = 255
        required_string_buffer_size = ctypes.c_size_t(0)
4714
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
4715
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
wxchan's avatar
wxchan committed
4716
        _safe_call(_LIB.LGBM_BoosterGetFeatureNames(
4717
            self._handle,
4718
            ctypes.c_int(num_feature),
wxchan's avatar
wxchan committed
4719
            ctypes.byref(tmp_out_len),
4720
            ctypes.c_size_t(reserved_string_buffer_size),
4721
            ctypes.byref(required_string_buffer_size),
wxchan's avatar
wxchan committed
4722
4723
4724
            ptr_string_buffers))
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
4725
4726
4727
4728
        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)]
4729
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
4730
            _safe_call(_LIB.LGBM_BoosterGetFeatureNames(
4731
                self._handle,
4732
4733
4734
4735
4736
                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))
4737
        return [string_buffers[i].value.decode('utf-8') for i in range(num_feature)]
wxchan's avatar
wxchan committed
4738

4739
4740
4741
4742
4743
    def feature_importance(
        self,
        importance_type: str = 'split',
        iteration: Optional[int] = None
    ) -> np.ndarray:
4744
        """Get feature importances.
4745

4746
4747
        Parameters
        ----------
4748
        importance_type : str, optional (default="split")
4749
4750
4751
            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.
4752
4753
4754
4755
        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).
4756

4757
4758
        Returns
        -------
4759
4760
        result : numpy array
            Array with feature importances.
4761
        """
4762
4763
        if iteration is None:
            iteration = self.best_iteration
4764
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4765
        result = np.empty(self.num_feature(), dtype=np.float64)
4766
        _safe_call(_LIB.LGBM_BoosterFeatureImportance(
4767
            self._handle,
4768
4769
4770
            ctypes.c_int(iteration),
            ctypes.c_int(importance_type_int),
            result.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
4771
        if importance_type_int == _C_API_FEATURE_IMPORTANCE_SPLIT:
4772
            return result.astype(np.int32)
4773
4774
        else:
            return result
4775

4776
4777
4778
4779
4780
4781
    def get_split_value_histogram(
        self,
        feature: Union[int, str],
        bins: Optional[Union[int, str]] = None,
        xgboost_style: bool = False
    ) -> Union[Tuple[np.ndarray, np.ndarray], np.ndarray, pd_DataFrame]:
4782
4783
4784
4785
        """Get split value histogram for the specified feature.

        Parameters
        ----------
4786
        feature : int or str
4787
4788
            The feature name or index the histogram is calculated for.
            If int, interpreted as index.
4789
            If str, interpreted as name.
4790

Nikita Titov's avatar
Nikita Titov committed
4791
4792
4793
            .. warning::

                Categorical features are not supported.
4794

4795
        bins : int, str or None, optional (default=None)
4796
4797
4798
            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.
4799
            If str, it should be one from the list of the supported values by ``numpy.histogram()`` function.
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
        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.
        """
4814
        def add(root: Dict[str, Any]) -> None:
4815
4816
            """Recursively add thresholds."""
            if 'split_index' in root:  # non-leaf
4817
                if feature_names is not None and isinstance(feature, str):
4818
4819
4820
4821
                    split_feature = feature_names[root['split_feature']]
                else:
                    split_feature = root['split_feature']
                if split_feature == feature:
4822
                    if isinstance(root['threshold'], str):
4823
4824
4825
                        raise LightGBMError('Cannot compute split value histogram for the categorical feature')
                    else:
                        values.append(root['threshold'])
4826
4827
4828
4829
4830
4831
                add(root['left_child'])
                add(root['right_child'])

        model = self.dump_model()
        feature_names = model.get('feature_names')
        tree_infos = model['tree_info']
4832
        values: List[float] = []
4833
4834
4835
        for tree_info in tree_infos:
            add(tree_info['tree_structure'])

4836
        if bins is None or isinstance(bins, int) and xgboost_style:
4837
4838
4839
4840
4841
4842
4843
            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:
4844
                return pd_DataFrame(ret, columns=['SplitValue', 'Count'])
4845
4846
4847
4848
4849
            else:
                return ret
        else:
            return hist, bin_edges

4850
4851
4852
4853
    def __inner_eval(
        self,
        data_name: str,
        data_idx: int,
4854
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]]
4855
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4856
        """Evaluate training or validation data."""
wxchan's avatar
wxchan committed
4857
        if data_idx >= self.__num_dataset:
4858
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
4859
4860
4861
        self.__get_eval_info()
        ret = []
        if self.__num_inner_eval > 0:
4862
            result = np.empty(self.__num_inner_eval, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
4863
            tmp_out_len = ctypes.c_int(0)
wxchan's avatar
wxchan committed
4864
            _safe_call(_LIB.LGBM_BoosterGetEval(
4865
                self._handle,
Guolin Ke's avatar
Guolin Ke committed
4866
                ctypes.c_int(data_idx),
wxchan's avatar
wxchan committed
4867
                ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
4868
                result.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
wxchan's avatar
wxchan committed
4869
            if tmp_out_len.value != self.__num_inner_eval:
4870
                raise ValueError("Wrong length of eval results")
4871
            for i in range(self.__num_inner_eval):
4872
4873
                ret.append((data_name, self.__name_inner_eval[i],
                            result[i], self.__higher_better_inner_eval[i]))
4874
4875
        if callable(feval):
            feval = [feval]
wxchan's avatar
wxchan committed
4876
4877
4878
4879
4880
        if feval is not None:
            if data_idx == 0:
                cur_data = self.train_set
            else:
                cur_data = self.valid_sets[data_idx - 1]
4881
4882
4883
4884
4885
4886
4887
4888
4889
            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
4890
4891
4892
                    ret.append((data_name, eval_name, val, is_higher_better))
        return ret

4893
    def __inner_predict(self, data_idx: int) -> np.ndarray:
4894
        """Predict for training and validation dataset."""
wxchan's avatar
wxchan committed
4895
        if data_idx >= self.__num_dataset:
4896
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
4897
4898
4899
4900
4901
        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
4902
            self.__inner_predict_buffer[data_idx] = np.empty(n_preds, dtype=np.float64)
4903
        # avoid to predict many time in one iteration
wxchan's avatar
wxchan committed
4904
4905
        if not self.__is_predicted_cur_iter[data_idx]:
            tmp_out_len = ctypes.c_int64(0)
4906
            data_ptr = self.__inner_predict_buffer[data_idx].ctypes.data_as(ctypes.POINTER(ctypes.c_double))  # type: ignore[union-attr]
wxchan's avatar
wxchan committed
4907
            _safe_call(_LIB.LGBM_BoosterGetPredict(
4908
                self._handle,
Guolin Ke's avatar
Guolin Ke committed
4909
                ctypes.c_int(data_idx),
wxchan's avatar
wxchan committed
4910
4911
                ctypes.byref(tmp_out_len),
                data_ptr))
4912
            if tmp_out_len.value != len(self.__inner_predict_buffer[data_idx]):  # type: ignore[arg-type]
4913
                raise ValueError(f"Wrong length of predict results for data {data_idx}")
wxchan's avatar
wxchan committed
4914
            self.__is_predicted_cur_iter[data_idx] = True
4915
        result: np.ndarray = self.__inner_predict_buffer[data_idx]  # type: ignore[assignment]
4916
4917
4918
4919
        if self.__num_class > 1:
            num_data = result.size // self.__num_class
            result = result.reshape(num_data, self.__num_class, order='F')
        return result
wxchan's avatar
wxchan committed
4920

4921
    def __get_eval_info(self) -> None:
4922
        """Get inner evaluation count and names."""
wxchan's avatar
wxchan committed
4923
4924
        if self.__need_reload_eval_info:
            self.__need_reload_eval_info = False
Guolin Ke's avatar
Guolin Ke committed
4925
            out_num_eval = ctypes.c_int(0)
4926
            # Get num of inner evals
wxchan's avatar
wxchan committed
4927
            _safe_call(_LIB.LGBM_BoosterGetEvalCounts(
4928
                self._handle,
wxchan's avatar
wxchan committed
4929
4930
4931
                ctypes.byref(out_num_eval)))
            self.__num_inner_eval = out_num_eval.value
            if self.__num_inner_eval > 0:
4932
                # Get name of eval metrics
Guolin Ke's avatar
Guolin Ke committed
4933
                tmp_out_len = ctypes.c_int(0)
4934
4935
4936
                reserved_string_buffer_size = 255
                required_string_buffer_size = ctypes.c_size_t(0)
                string_buffers = [
4937
                    ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(self.__num_inner_eval)
4938
                ]
4939
                ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
wxchan's avatar
wxchan committed
4940
                _safe_call(_LIB.LGBM_BoosterGetEvalNames(
4941
                    self._handle,
4942
                    ctypes.c_int(self.__num_inner_eval),
wxchan's avatar
wxchan committed
4943
                    ctypes.byref(tmp_out_len),
4944
                    ctypes.c_size_t(reserved_string_buffer_size),
4945
                    ctypes.byref(required_string_buffer_size),
wxchan's avatar
wxchan committed
4946
4947
                    ptr_string_buffers))
                if self.__num_inner_eval != tmp_out_len.value:
4948
                    raise ValueError("Length of eval names doesn't equal with num_evals")
4949
4950
4951
4952
4953
4954
                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)
                    ]
4955
                    ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))  # type: ignore[misc]
4956
                    _safe_call(_LIB.LGBM_BoosterGetEvalNames(
4957
                        self._handle,
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
                        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)
                ]
                self.__higher_better_inner_eval = [
                    name.startswith(('auc', 'ndcg@', 'map@', 'average_precision')) for name in self.__name_inner_eval
                ]