basic.py 179 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 json
wxchan's avatar
wxchan committed
6
import warnings
7
from collections import OrderedDict
8
from copy import deepcopy
9
from enum import Enum
10
from functools import wraps
11
from os import SEEK_END, environ
12
13
from os.path import getsize
from pathlib import Path
14
from tempfile import NamedTemporaryFile
15
from typing import Any, Callable, Dict, Iterable, List, Optional, Set, Tuple, Union
wxchan's avatar
wxchan committed
16
17
18
19

import numpy as np
import scipy.sparse

20
from .compat import PANDAS_INSTALLED, concat, dt_DataTable, pd_CategoricalDtype, pd_DataFrame, pd_Series
wxchan's avatar
wxchan committed
21
22
from .libpath import find_lib_path

23
24
25
26
27
28
29
30
31
__all__ = [
    'Booster',
    'Dataset',
    'LGBMDeprecationWarning',
    'LightGBMError',
    'register_logger',
    'Sequence',
]

32
_DatasetHandle = ctypes.c_void_p
33
_LGBM_EvalFunctionResultType = Tuple[str, float, bool]
34
_LGBM_BoosterBestScoreType = Dict[str, Dict[str, float]]
35
_LGBM_BoosterEvalMethodResultType = Tuple[str, str, float, bool]
36
37
_LGBM_CategoricalFeatureConfiguration = Union[List[str], List[int], str]
_LGBM_FeatureNameConfiguration = Union[List[str], str]
38
39
40
41
42
43
44
45
46
47
48
_LGBM_TrainDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
    scipy.sparse.spmatrix,
    "Sequence",
    List["Sequence"],
    List[np.ndarray]
]
49
50
51
52
53
54
_LGBM_LabelType = Union[
    list,
    np.ndarray,
    pd_Series,
    pd_DataFrame
]
55
56
57
58
59
60
61
62
_LGBM_PredictDataType = Union[
    str,
    Path,
    np.ndarray,
    pd_DataFrame,
    dt_DataTable,
    scipy.sparse.spmatrix
]
63

64
65
66
ZERO_THRESHOLD = 1e-35


67
68
69
70
def _is_zero(x: float) -> bool:
    return -ZERO_THRESHOLD <= x <= ZERO_THRESHOLD


71
def _get_sample_count(total_nrow: int, params: str) -> int:
72
73
74
    sample_cnt = ctypes.c_int(0)
    _safe_call(_LIB.LGBM_GetSampleCount(
        ctypes.c_int32(total_nrow),
75
        _c_str(params),
76
77
78
79
        ctypes.byref(sample_cnt),
    ))
    return sample_cnt.value

wxchan's avatar
wxchan committed
80

81
82
83
84
85
86
class _MissingType(Enum):
    NONE = 'None'
    NAN = 'NaN'
    ZERO = 'Zero'


87
class _DummyLogger:
88
    def info(self, msg: str) -> None:
89
90
        print(msg)

91
    def warning(self, msg: str) -> None:
92
93
94
        warnings.warn(msg, stacklevel=3)


95
96
97
_LOGGER: Any = _DummyLogger()
_INFO_METHOD_NAME = "info"
_WARNING_METHOD_NAME = "warning"
98
99


100
101
102
103
def _has_method(logger: Any, method_name: str) -> bool:
    return callable(getattr(logger, method_name, None))


104
105
106
def register_logger(
    logger: Any, info_method_name: str = "info", warning_method_name: str = "warning"
) -> None:
107
108
109
110
    """Register custom logger.

    Parameters
    ----------
111
    logger : Any
112
        Custom logger.
113
114
115
116
    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.
117
    """
118
119
120
121
122
123
    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
124
    _LOGGER = logger
125
126
    _INFO_METHOD_NAME = info_method_name
    _WARNING_METHOD_NAME = warning_method_name
127
128


129
def _normalize_native_string(func: Callable[[str], None]) -> Callable[[str], None]:
130
    """Join log messages from native library which come by chunks."""
131
    msg_normalized: List[str] = []
132
133

    @wraps(func)
134
    def wrapper(msg: str) -> None:
135
136
137
138
139
140
141
142
143
144
145
        nonlocal msg_normalized
        if msg.strip() == '':
            msg = ''.join(msg_normalized)
            msg_normalized = []
            return func(msg)
        else:
            msg_normalized.append(msg)

    return wrapper


146
def _log_info(msg: str) -> None:
147
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
148
149


150
def _log_warning(msg: str) -> None:
151
    getattr(_LOGGER, _WARNING_METHOD_NAME)(msg)
152
153
154


@_normalize_native_string
155
def _log_native(msg: str) -> None:
156
    getattr(_LOGGER, _INFO_METHOD_NAME)(msg)
157
158


159
def _log_callback(msg: bytes) -> None:
160
    """Redirect logs from native library into Python."""
161
    _log_native(str(msg.decode('utf-8')))
162
163


164
def _load_lib() -> ctypes.CDLL:
165
    """Load LightGBM library."""
wxchan's avatar
wxchan committed
166
167
168
    lib_path = find_lib_path()
    lib = ctypes.cdll.LoadLibrary(lib_path[0])
    lib.LGBM_GetLastError.restype = ctypes.c_char_p
169
    callback = ctypes.CFUNCTYPE(None, ctypes.c_char_p)
170
    lib.callback = callback(_log_callback)  # type: ignore[attr-defined]
171
    if lib.LGBM_RegisterLogCallback(lib.callback) != 0:
172
        raise LightGBMError(lib.LGBM_GetLastError().decode('utf-8'))
wxchan's avatar
wxchan committed
173
174
    return lib

wxchan's avatar
wxchan committed
175

176
177
178
179
180
181
182
# 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
183

wxchan's avatar
wxchan committed
184

185
_NUMERIC_TYPES = (int, float, bool)
186
_ArrayLike = Union[List, np.ndarray, pd_Series]
187
188


189
def _safe_call(ret: int) -> None:
190
191
    """Check the return value from C API call.

wxchan's avatar
wxchan committed
192
193
194
    Parameters
    ----------
    ret : int
195
        The return value from C API calls.
wxchan's avatar
wxchan committed
196
197
    """
    if ret != 0:
198
        raise LightGBMError(_LIB.LGBM_GetLastError().decode('utf-8'))
wxchan's avatar
wxchan committed
199

wxchan's avatar
wxchan committed
200

201
def _is_numeric(obj: Any) -> bool:
202
    """Check whether object is a number or not, include numpy number, etc."""
wxchan's avatar
wxchan committed
203
204
205
    try:
        float(obj)
        return True
wxchan's avatar
wxchan committed
206
207
208
    except (TypeError, ValueError):
        # TypeError: obj is not a string or a number
        # ValueError: invalid literal
wxchan's avatar
wxchan committed
209
210
        return False

wxchan's avatar
wxchan committed
211

212
def _is_numpy_1d_array(data: Any) -> bool:
213
    """Check whether data is a numpy 1-D array."""
214
    return isinstance(data, np.ndarray) and len(data.shape) == 1
wxchan's avatar
wxchan committed
215

wxchan's avatar
wxchan committed
216

217
def _is_numpy_column_array(data: Any) -> bool:
218
219
220
221
222
223
224
    """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


225
def _cast_numpy_array_to_dtype(array: np.ndarray, dtype: np.dtype) -> np.ndarray:
226
    """Cast numpy array to given dtype."""
227
228
229
230
231
    if array.dtype == dtype:
        return array
    return array.astype(dtype=dtype, copy=False)


232
def _is_1d_list(data: Any) -> bool:
233
    """Check whether data is a 1-D list."""
234
    return isinstance(data, list) and (not data or _is_numeric(data[0]))
wxchan's avatar
wxchan committed
235

wxchan's avatar
wxchan committed
236

237
238
239
def _is_1d_collection(data: Any) -> bool:
    """Check whether data is a 1-D collection."""
    return (
240
        _is_numpy_1d_array(data)
241
        or _is_numpy_column_array(data)
242
        or _is_1d_list(data)
243
244
245
246
        or isinstance(data, pd_Series)
    )


247
def _list_to_1d_numpy(data, dtype=np.float32, name='list'):
248
    """Convert data to numpy 1-D array."""
249
    if _is_numpy_1d_array(data):
250
        return _cast_numpy_array_to_dtype(data, dtype)
251
    elif _is_numpy_column_array(data):
252
253
        _log_warning('Converting column-vector to 1d array')
        array = data.ravel()
254
        return _cast_numpy_array_to_dtype(array, dtype)
255
    elif _is_1d_list(data):
wxchan's avatar
wxchan committed
256
        return np.array(data, dtype=dtype, copy=False)
257
    elif isinstance(data, pd_Series):
258
        _check_for_bad_pandas_dtypes(data.to_frame().dtypes)
259
        return np.array(data, dtype=dtype, copy=False)  # SparseArray should be supported as well
wxchan's avatar
wxchan committed
260
    else:
261
262
        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
263

wxchan's avatar
wxchan committed
264

265
266
267
268
269
270
271
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."""
272
    return isinstance(data, list) and len(data) > 0 and _is_1d_list(data[0])
273
274
275
276
277
278
279
280
281
282
283
284
285
286


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)
    )


def _data_to_2d_numpy(data: Any, dtype: type = np.float32, name: str = 'list') -> np.ndarray:
    """Convert data to numpy 2-D array."""
    if _is_numpy_2d_array(data):
287
        return _cast_numpy_array_to_dtype(data, dtype)
288
289
290
    if _is_2d_list(data):
        return np.array(data, dtype=dtype)
    if isinstance(data, pd_DataFrame):
291
        _check_for_bad_pandas_dtypes(data.dtypes)
292
        return _cast_numpy_array_to_dtype(data.values, dtype)
293
294
295
296
    raise TypeError(f"Wrong type({type(data).__name__}) for {name}.\n"
                    "It should be list of lists, numpy 2-D array or pandas DataFrame")


297
def _cfloat32_array_to_numpy(cptr: "ctypes._Pointer", length: int) -> np.ndarray:
298
    """Convert a ctypes float pointer array to a numpy array."""
wxchan's avatar
wxchan committed
299
    if isinstance(cptr, ctypes.POINTER(ctypes.c_float)):
300
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
301
    else:
302
        raise RuntimeError('Expected float pointer')
wxchan's avatar
wxchan committed
303

Guolin Ke's avatar
Guolin Ke committed
304

305
def _cfloat64_array_to_numpy(cptr: "ctypes._Pointer", length: int) -> np.ndarray:
306
    """Convert a ctypes double pointer array to a numpy array."""
Guolin Ke's avatar
Guolin Ke committed
307
    if isinstance(cptr, ctypes.POINTER(ctypes.c_double)):
308
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
Guolin Ke's avatar
Guolin Ke committed
309
310
311
    else:
        raise RuntimeError('Expected double pointer')

wxchan's avatar
wxchan committed
312

313
def _cint32_array_to_numpy(cptr: "ctypes._Pointer", length: int) -> np.ndarray:
314
    """Convert a ctypes int pointer array to a numpy array."""
wxchan's avatar
wxchan committed
315
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int32)):
316
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
wxchan's avatar
wxchan committed
317
    else:
318
319
320
        raise RuntimeError('Expected int32 pointer')


321
def _cint64_array_to_numpy(cptr: "ctypes._Pointer", length: int) -> np.ndarray:
322
323
    """Convert a ctypes int pointer array to a numpy array."""
    if isinstance(cptr, ctypes.POINTER(ctypes.c_int64)):
324
        return np.ctypeslib.as_array(cptr, shape=(length,)).copy()
325
326
    else:
        raise RuntimeError('Expected int64 pointer')
wxchan's avatar
wxchan committed
327

wxchan's avatar
wxchan committed
328

329
def _c_str(string: str) -> ctypes.c_char_p:
330
    """Convert a Python string to C string."""
wxchan's avatar
wxchan committed
331
332
    return ctypes.c_char_p(string.encode('utf-8'))

wxchan's avatar
wxchan committed
333

334
def _c_array(ctype: type, values: List[Any]) -> ctypes.Array:
335
    """Convert a Python array to C array."""
336
    return (ctype * len(values))(*values)  # type: ignore[operator]
wxchan's avatar
wxchan committed
337

wxchan's avatar
wxchan committed
338

339
def _json_default_with_numpy(obj: Any) -> Any:
340
341
342
343
344
345
346
347
348
    """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


349
350
351
352
353
354
355
356
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)


357
def _param_dict_to_str(data: Optional[Dict[str, Any]]) -> str:
358
    """Convert Python dictionary to string, which is passed to C API."""
359
    if data is None or not data:
wxchan's avatar
wxchan committed
360
361
362
        return ""
    pairs = []
    for key, val in data.items():
363
        if isinstance(val, (list, tuple, set)) or _is_numpy_1d_array(val):
364
            pairs.append(f"{key}={','.join(map(_to_string, val))}")
365
        elif isinstance(val, (str, Path, _NUMERIC_TYPES)) or _is_numeric(val):
366
            pairs.append(f"{key}={val}")
367
        elif val is not None:
368
            raise TypeError(f'Unknown type of parameter:{key}, got:{type(val).__name__}')
wxchan's avatar
wxchan committed
369
    return ' '.join(pairs)
370

wxchan's avatar
wxchan committed
371

372
class _TempFile:
373
374
    """Proxy class to workaround errors on Windows."""

375
376
377
    def __enter__(self):
        with NamedTemporaryFile(prefix="lightgbm_tmp_", delete=True) as f:
            self.name = f.name
378
            self.path = Path(self.name)
379
        return self
wxchan's avatar
wxchan committed
380

381
    def __exit__(self, exc_type, exc_val, exc_tb):
382
383
        if self.path.is_file():
            self.path.unlink()
384

wxchan's avatar
wxchan committed
385

386
class LightGBMError(Exception):
387
388
    """Error thrown by LightGBM."""

389
390
391
    pass


392
393
394
395
396
397
398
399
# DeprecationWarning is not shown by default, so let's create our own with higher level
class LGBMDeprecationWarning(UserWarning):
    """Custom deprecation warning."""

    pass


class _ConfigAliases:
400
401
402
403
    # lazy evaluation to allow import without dynamic library, e.g., for docs generation
    aliases = None

    @staticmethod
404
    def _get_all_param_aliases() -> Dict[str, List[str]]:
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
        buffer_len = 1 << 20
        tmp_out_len = ctypes.c_int64(0)
        string_buffer = ctypes.create_string_buffer(buffer_len)
        ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
        _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)
            ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
            _safe_call(_LIB.LGBM_DumpParamAliases(
                ctypes.c_int64(actual_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
        aliases = json.loads(
            string_buffer.value.decode('utf-8'),
424
            object_hook=lambda obj: {k: [k] + v for k, v in obj.items()}
425
426
        )
        return aliases
427
428

    @classmethod
429
430
431
    def get(cls, *args) -> Set[str]:
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
432
433
        ret = set()
        for i in args:
434
            ret.update(cls.get_sorted(i))
435
436
        return ret

437
438
439
440
441
442
    @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])

443
    @classmethod
444
445
446
    def get_by_alias(cls, *args) -> Set[str]:
        if cls.aliases is None:
            cls.aliases = cls._get_all_param_aliases()
447
448
449
450
        ret = set(args)
        for arg in args:
            for aliases in cls.aliases.values():
                if arg in aliases:
451
                    ret.update(aliases)
452
453
454
                    break
        return ret

455

456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
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)

477
478
    aliases = _ConfigAliases.get_sorted(main_param_name)
    aliases = [a for a in aliases if a != main_param_name]
479
480

    # if main_param_name was provided, keep that value and remove all aliases
481
    if main_param_name in params.keys():
482
483
484
        for param in aliases:
            params.pop(param, None)
        return params
485

486
487
488
489
490
    # 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
491

492
493
494
495
496
497
498
    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
499
500
501
502

    return params


503
_MAX_INT32 = (1 << 31) - 1
504

505
"""Macro definition of data type in C API of LightGBM"""
506
507
508
509
_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
510

511
"""Matrix is row major in Python"""
512
_C_API_IS_ROW_MAJOR = 1
wxchan's avatar
wxchan committed
513

514
"""Macro definition of prediction type in C API of LightGBM"""
515
516
517
518
_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
519

520
"""Macro definition of sparse matrix type"""
521
522
_C_API_MATRIX_TYPE_CSR = 0
_C_API_MATRIX_TYPE_CSC = 1
523

524
"""Macro definition of feature importance type"""
525
526
_C_API_FEATURE_IMPORTANCE_SPLIT = 0
_C_API_FEATURE_IMPORTANCE_GAIN = 1
527

528
"""Data type of data field"""
529
530
531
532
533
534
_FIELD_TYPE_MAPPER = {
    "label": _C_API_DTYPE_FLOAT32,
    "weight": _C_API_DTYPE_FLOAT32,
    "init_score": _C_API_DTYPE_FLOAT64,
    "group": _C_API_DTYPE_INT32
}
wxchan's avatar
wxchan committed
535

536
"""String name to int feature importance type mapper"""
537
538
539
540
_FEATURE_IMPORTANCE_TYPE_MAPPER = {
    "split": _C_API_FEATURE_IMPORTANCE_SPLIT,
    "gain": _C_API_FEATURE_IMPORTANCE_GAIN
}
541

wxchan's avatar
wxchan committed
542

543
def _convert_from_sliced_object(data: np.ndarray) -> np.ndarray:
544
    """Fix the memory of multi-dimensional sliced object."""
545
    if isinstance(data, np.ndarray) and isinstance(data.base, np.ndarray):
546
        if not data.flags.c_contiguous:
547
548
            _log_warning("Usage of np.ndarray subset (sliced data) is not recommended "
                         "due to it will double the peak memory cost in LightGBM.")
549
550
551
552
            return np.copy(data)
    return data


553
def _c_float_array(data):
554
    """Get pointer of float numpy array / list."""
555
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
556
        data = np.array(data, copy=False)
557
    if _is_numpy_1d_array(data):
558
        data = _convert_from_sliced_object(data)
559
        assert data.flags.c_contiguous
wxchan's avatar
wxchan committed
560
561
        if data.dtype == np.float32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
562
            type_data = _C_API_DTYPE_FLOAT32
wxchan's avatar
wxchan committed
563
564
        elif data.dtype == np.float64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_double))
565
            type_data = _C_API_DTYPE_FLOAT64
wxchan's avatar
wxchan committed
566
        else:
567
            raise TypeError(f"Expected np.float32 or np.float64, met type({data.dtype})")
wxchan's avatar
wxchan committed
568
    else:
569
        raise TypeError(f"Unknown type({type(data).__name__})")
570
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
571

wxchan's avatar
wxchan committed
572

573
def _c_int_array(data):
574
    """Get pointer of int numpy array / list."""
575
    if _is_1d_list(data):
wxchan's avatar
wxchan committed
576
        data = np.array(data, copy=False)
577
    if _is_numpy_1d_array(data):
578
        data = _convert_from_sliced_object(data)
579
        assert data.flags.c_contiguous
wxchan's avatar
wxchan committed
580
581
        if data.dtype == np.int32:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int32))
582
            type_data = _C_API_DTYPE_INT32
wxchan's avatar
wxchan committed
583
584
        elif data.dtype == np.int64:
            ptr_data = data.ctypes.data_as(ctypes.POINTER(ctypes.c_int64))
585
            type_data = _C_API_DTYPE_INT64
wxchan's avatar
wxchan committed
586
        else:
587
            raise TypeError(f"Expected np.int32 or np.int64, met type({data.dtype})")
wxchan's avatar
wxchan committed
588
    else:
589
        raise TypeError(f"Unknown type({type(data).__name__})")
590
    return (ptr_data, type_data, data)  # return `data` to avoid the temporary copy is freed
wxchan's avatar
wxchan committed
591

wxchan's avatar
wxchan committed
592

593
def _is_allowed_numpy_dtype(dtype) -> bool:
594
    float128 = getattr(np, 'float128', type(None))
595
596
597
598
    return (
        issubclass(dtype, (np.integer, np.floating, np.bool_))
        and not issubclass(dtype, (np.timedelta64, float128))
    )
599
600


601
def _check_for_bad_pandas_dtypes(pandas_dtypes_series: pd_Series) -> None:
602
603
    bad_pandas_dtypes = [
        f'{column_name}: {pandas_dtype}'
604
        for column_name, pandas_dtype in pandas_dtypes_series.items()
605
        if not _is_allowed_numpy_dtype(pandas_dtype.type)
606
607
608
609
    ]
    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)}')
610
611


612
613
614
615
616
617
def _data_from_pandas(
    data,
    feature_name: Optional[_LGBM_FeatureNameConfiguration],
    categorical_feature: Optional[_LGBM_CategoricalFeatureConfiguration],
    pandas_categorical: Optional[List[List]]
):
618
    if isinstance(data, pd_DataFrame):
619
620
        if len(data.shape) != 2 or data.shape[0] < 1:
            raise ValueError('Input data must be 2 dimensional and non empty.')
621
        if feature_name == 'auto' or feature_name is None:
622
            data = data.rename(columns=str, copy=False)
623
        cat_cols = [col for col, dtype in zip(data.columns, data.dtypes) if isinstance(dtype, pd_CategoricalDtype)]
624
        cat_cols_not_ordered = [col for col in cat_cols if not data[col].cat.ordered]
625
626
627
628
629
        if pandas_categorical is None:  # train dataset
            pandas_categorical = [list(data[col].cat.categories) for col in cat_cols]
        else:
            if len(cat_cols) != len(pandas_categorical):
                raise ValueError('train and valid dataset categorical_feature do not match.')
630
            for col, category in zip(cat_cols, pandas_categorical):
631
632
                if list(data[col].cat.categories) != list(category):
                    data[col] = data[col].cat.set_categories(category)
633
        if len(cat_cols):  # cat_cols is list
634
            data = data.copy(deep=False)  # not alter origin DataFrame
635
            data[cat_cols] = data[cat_cols].apply(lambda x: x.cat.codes).replace({-1: np.nan})
636
637
638
        if categorical_feature is not None:
            if feature_name is None:
                feature_name = list(data.columns)
639
            if categorical_feature == 'auto':  # use cat cols from DataFrame
640
                categorical_feature = cat_cols_not_ordered
641
642
            else:  # use cat cols specified by user
                categorical_feature = list(categorical_feature)
643
644
        if feature_name == 'auto':
            feature_name = list(data.columns)
645
        _check_for_bad_pandas_dtypes(data.dtypes)
646
647
648
        df_dtypes = [dtype.type for dtype in data.dtypes]
        df_dtypes.append(np.float32)  # so that the target dtype considers floats
        target_dtype = np.find_common_type(df_dtypes, [])
649
650
651
652
653
654
655
656
657
658
        try:
            # most common case (no nullable dtypes)
            data = 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
            data = 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
            data = data.to_numpy(dtype=target_dtype, na_value=np.nan)
659
660
661
662
663
664
    else:
        if feature_name == 'auto':
            feature_name = None
        if categorical_feature == 'auto':
            categorical_feature = None
    return data, feature_name, categorical_feature, pandas_categorical
665
666


667
668
669
670
def _dump_pandas_categorical(
    pandas_categorical: Optional[List[List]],
    file_name: Optional[Union[str, Path]] = None
) -> str:
671
    categorical_json = json.dumps(pandas_categorical, default=_json_default_with_numpy)
672
    pandas_str = f'\npandas_categorical:{categorical_json}\n'
673
674
675
676
677
678
    if file_name is not None:
        with open(file_name, 'a') as f:
            f.write(pandas_str)
    return pandas_str


679
680
681
def _load_pandas_categorical(
    file_name: Optional[Union[str, Path]] = None,
    model_str: Optional[str] = None
682
) -> Optional[List[List]]:
683
684
    pandas_key = 'pandas_categorical:'
    offset = -len(pandas_key)
685
    if file_name is not None:
686
        max_offset = -getsize(file_name)
687
688
689
690
        with open(file_name, 'rb') as f:
            while True:
                if offset < max_offset:
                    offset = max_offset
691
                f.seek(offset, SEEK_END)
692
693
694
695
                lines = f.readlines()
                if len(lines) >= 2:
                    break
                offset *= 2
696
        last_line = lines[-1].decode('utf-8').strip()
697
        if not last_line.startswith(pandas_key):
698
            last_line = lines[-2].decode('utf-8').strip()
699
    elif model_str is not None:
700
701
702
703
704
705
        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
706
707


708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
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**.

728
729
    .. versionadded:: 3.3.0

730
731
732
733
734
735
736
737
738
    Attributes
    ----------
    batch_size : int
        Default size of a batch.
    """

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

    @abc.abstractmethod
739
    def __getitem__(self, idx: Union[int, slice, List[int]]) -> np.ndarray:
740
741
742
743
744
745
746
        """Return data for given row index.

        A basic implementation should look like this:

        .. code-block:: python

            if isinstance(idx, numbers.Integral):
747
                return self._get_one_line(idx)
748
            elif isinstance(idx, slice):
749
750
                return np.stack([self._get_one_line(i) for i in range(idx.start, idx.stop)])
            elif isinstance(idx, list):
751
                # Only required if using ``Dataset.subset()``.
752
                return np.array([self._get_one_line(i) for i in idx])
753
            else:
754
                raise TypeError(f"Sequence index must be integer, slice or list, got {type(idx).__name__}")
755
756
757

        Parameters
        ----------
758
        idx : int, slice[int], list[int]
759
760
761
762
            Item index.

        Returns
        -------
763
        result : numpy 1-D array or numpy 2-D array
764
            1-D array if idx is int, 2-D array if idx is slice or list.
765
766
767
768
769
770
771
772
773
        """
        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__()")


774
class _InnerPredictor:
775
776
777
778
779
    """_InnerPredictor of LightGBM.

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

Nikita Titov's avatar
Nikita Titov committed
780
781
782
    .. note::

        Can be converted from Booster, but cannot be converted to Booster.
Guolin Ke's avatar
Guolin Ke committed
783
    """
784

785
786
787
788
789
790
    def __init__(
        self,
        model_file: Optional[Union[str, Path]] = None,
        booster_handle: Optional[ctypes.c_void_p] = None,
        pred_parameter: Optional[Dict[str, Any]] = None
    ):
791
        """Initialize the _InnerPredictor.
wxchan's avatar
wxchan committed
792
793
794

        Parameters
        ----------
795
        model_file : str, pathlib.Path or None, optional (default=None)
wxchan's avatar
wxchan committed
796
            Path to the model file.
797
798
799
        booster_handle : object or None, optional (default=None)
            Handle of Booster.
        pred_parameter: dict or None, optional (default=None)
800
            Other parameters for the prediction.
wxchan's avatar
wxchan committed
801
802
803
804
805
        """
        self.handle = ctypes.c_void_p()
        self.__is_manage_handle = True
        if model_file is not None:
            """Prediction task"""
Guolin Ke's avatar
Guolin Ke committed
806
            out_num_iterations = ctypes.c_int(0)
wxchan's avatar
wxchan committed
807
            _safe_call(_LIB.LGBM_BoosterCreateFromModelfile(
808
                _c_str(str(model_file)),
wxchan's avatar
wxchan committed
809
810
                ctypes.byref(out_num_iterations),
                ctypes.byref(self.handle)))
Guolin Ke's avatar
Guolin Ke committed
811
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
812
813
814
815
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
                self.handle,
                ctypes.byref(out_num_class)))
            self.num_class = out_num_class.value
816
            self.num_total_iteration = out_num_iterations.value
817
            self.pandas_categorical = _load_pandas_categorical(file_name=model_file)
wxchan's avatar
wxchan committed
818
        elif booster_handle is not None:
Guolin Ke's avatar
Guolin Ke committed
819
            self.__is_manage_handle = False
wxchan's avatar
wxchan committed
820
            self.handle = booster_handle
Guolin Ke's avatar
Guolin Ke committed
821
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
822
823
824
825
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
                self.handle,
                ctypes.byref(out_num_class)))
            self.num_class = out_num_class.value
826
            self.num_total_iteration = self.current_iteration()
827
            self.pandas_categorical = None
wxchan's avatar
wxchan committed
828
        else:
829
            raise TypeError('Need model_file or booster_handle to create a predictor')
wxchan's avatar
wxchan committed
830

831
        pred_parameter = {} if pred_parameter is None else pred_parameter
832
        self.pred_parameter = _param_dict_to_str(pred_parameter)
cbecker's avatar
cbecker committed
833

834
    def __del__(self) -> None:
835
836
837
838
839
        try:
            if self.__is_manage_handle:
                _safe_call(_LIB.LGBM_BoosterFree(self.handle))
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
840

841
    def __getstate__(self) -> Dict[str, Any]:
842
843
844
845
        this = self.__dict__.copy()
        this.pop('handle', None)
        return this

846
847
    def predict(
        self,
848
        data: _LGBM_PredictDataType,
849
850
851
852
853
854
855
        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
856
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
857
        """Predict logic.
wxchan's avatar
wxchan committed
858
859
860

        Parameters
        ----------
861
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame or scipy.sparse
862
            Data source for prediction.
863
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
864
865
        start_iteration : int, optional (default=0)
            Start index of the iteration to predict.
866
867
868
869
870
871
872
873
874
875
876
        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.
877
878
879
        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
880
881
882

        Returns
        -------
883
        result : numpy array, scipy.sparse or list of scipy.sparse
884
            Prediction result.
885
            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
886
        """
wxchan's avatar
wxchan committed
887
        if isinstance(data, Dataset):
888
            raise TypeError("Cannot use Dataset instance for prediction, please use raw data instead")
889
890
891
892
893
894
895
896
897
898
899
        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(
                    self.handle,
                    ptr_names,
                    ctypes.c_int(len(data_names)),
                )
            )
900
        data = _data_from_pandas(data, None, None, self.pandas_categorical)[0]
901
        predict_type = _C_API_PREDICT_NORMAL
wxchan's avatar
wxchan committed
902
        if raw_score:
903
            predict_type = _C_API_PREDICT_RAW_SCORE
wxchan's avatar
wxchan committed
904
        if pred_leaf:
905
            predict_type = _C_API_PREDICT_LEAF_INDEX
906
        if pred_contrib:
907
            predict_type = _C_API_PREDICT_CONTRIB
wxchan's avatar
wxchan committed
908
        int_data_has_header = 1 if data_has_header else 0
cbecker's avatar
cbecker committed
909

910
        if isinstance(data, (str, Path)):
911
            with _TempFile() as f:
wxchan's avatar
wxchan committed
912
913
                _safe_call(_LIB.LGBM_BoosterPredictForFile(
                    self.handle,
914
                    _c_str(str(data)),
Guolin Ke's avatar
Guolin Ke committed
915
916
                    ctypes.c_int(int_data_has_header),
                    ctypes.c_int(predict_type),
917
                    ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
918
                    ctypes.c_int(num_iteration),
919
920
                    _c_str(self.pred_parameter),
                    _c_str(f.name)))
921
922
                preds = np.loadtxt(f.name, dtype=np.float64)
                nrow = preds.shape[0]
wxchan's avatar
wxchan committed
923
        elif isinstance(data, scipy.sparse.csr_matrix):
924
925
926
927
928
929
            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
930
        elif isinstance(data, scipy.sparse.csc_matrix):
931
932
933
934
935
936
            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
937
        elif isinstance(data, np.ndarray):
938
939
940
941
942
943
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
944
945
946
        elif isinstance(data, list):
            try:
                data = np.array(data)
947
            except BaseException:
948
                raise ValueError('Cannot convert data list to numpy array.')
949
950
951
952
953
954
            preds, nrow = self.__pred_for_np2d(
                mat=data,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
955
        elif isinstance(data, dt_DataTable):
956
957
958
959
960
961
            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
962
963
        else:
            try:
964
                _log_warning('Converting data to scipy sparse matrix.')
wxchan's avatar
wxchan committed
965
                csr = scipy.sparse.csr_matrix(data)
966
            except BaseException:
967
                raise TypeError(f'Cannot predict data for type {type(data).__name__}')
968
969
970
971
972
973
            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
974
975
        if pred_leaf:
            preds = preds.astype(np.int32)
976
        is_sparse = scipy.sparse.issparse(preds) or isinstance(preds, list)
977
        if not is_sparse and preds.size != nrow:
wxchan's avatar
wxchan committed
978
            if preds.size % nrow == 0:
979
                preds = preds.reshape(nrow, -1)
wxchan's avatar
wxchan committed
980
            else:
981
                raise ValueError(f'Length of predict result ({preds.size}) cannot be divide nrow ({nrow})')
wxchan's avatar
wxchan committed
982
983
        return preds

984
985
986
987
988
989
990
    def __get_num_preds(
        self,
        start_iteration: int,
        num_iteration: int,
        nrow: int,
        predict_type: int
    ) -> int:
991
        """Get size of prediction result."""
992
        if nrow > _MAX_INT32:
993
            raise LightGBMError('LightGBM cannot perform prediction for data '
994
                                f'with number of rows greater than MAX_INT32 ({_MAX_INT32}).\n'
995
                                'You can split your data into chunks '
996
                                'and then concatenate predictions for them')
Guolin Ke's avatar
Guolin Ke committed
997
998
999
        n_preds = ctypes.c_int64(0)
        _safe_call(_LIB.LGBM_BoosterCalcNumPredict(
            self.handle,
Guolin Ke's avatar
Guolin Ke committed
1000
1001
            ctypes.c_int(nrow),
            ctypes.c_int(predict_type),
1002
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
1003
            ctypes.c_int(num_iteration),
Guolin Ke's avatar
Guolin Ke committed
1004
1005
            ctypes.byref(n_preds)))
        return n_preds.value
wxchan's avatar
wxchan committed
1006

1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
    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)
1020
1021
1022
1023
1024
1025
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=mat.shape[0],
            predict_type=predict_type
        )
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
        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(
            self.handle,
            ptr_data,
            ctypes.c_int(type_ptr_data),
            ctypes.c_int32(mat.shape[0]),
            ctypes.c_int32(mat.shape[1]),
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
            ctypes.c_int(predict_type),
            ctypes.c_int(start_iteration),
            ctypes.c_int(num_iteration),
            _c_str(self.pred_parameter),
            ctypes.byref(out_num_preds),
            preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
        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]:
1055
        """Predict for a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
1056
        if len(mat.shape) != 2:
1057
            raise ValueError('Input numpy.ndarray or list must be 2 dimensional')
wxchan's avatar
wxchan committed
1058

1059
        nrow = mat.shape[0]
1060
1061
        if nrow > _MAX_INT32:
            sections = np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)
1062
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1063
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff([0] + list(sections) + [nrow])]
1064
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1065
            preds = np.empty(sum(n_preds), dtype=np.float64)
1066
1067
            for chunk, (start_idx_pred, end_idx_pred) in zip(np.array_split(mat, sections),
                                                             zip(n_preds_sections, n_preds_sections[1:])):
1068
                # avoid memory consumption by arrays concatenation operations
1069
1070
1071
1072
1073
1074
1075
                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]
                )
1076
            return preds, nrow
wxchan's avatar
wxchan committed
1077
        else:
1078
1079
1080
1081
1082
1083
1084
            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
1085

1086
1087
1088
    def __create_sparse_native(
        self,
        cs: Union[scipy.sparse.csc_matrix, scipy.sparse.csr_matrix],
1089
1090
1091
1092
1093
1094
        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,
1095
        is_csr: bool
1096
    ) -> Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]]:
1097
1098
1099
        # create numpy array from output arrays
        data_indices_len = out_shape[0]
        indptr_len = out_shape[1]
1100
        if indptr_type == _C_API_DTYPE_INT32:
1101
            out_indptr = _cint32_array_to_numpy(out_ptr_indptr, indptr_len)
1102
        elif indptr_type == _C_API_DTYPE_INT64:
1103
            out_indptr = _cint64_array_to_numpy(out_ptr_indptr, indptr_len)
1104
1105
        else:
            raise TypeError("Expected int32 or int64 type for indptr")
1106
        if data_type == _C_API_DTYPE_FLOAT32:
1107
            out_data = _cfloat32_array_to_numpy(out_ptr_data, data_indices_len)
1108
        elif data_type == _C_API_DTYPE_FLOAT64:
1109
            out_data = _cfloat64_array_to_numpy(out_ptr_data, data_indices_len)
1110
1111
        else:
            raise TypeError("Expected float32 or float64 type for data")
1112
        out_indices = _cint32_array_to_numpy(out_ptr_indices, data_indices_len)
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
        # 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

1141
1142
1143
1144
1145
1146
1147
1148
1149
    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
1150
1151
1152
1153
1154
1155
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
            predict_type=predict_type
        )
1156
1157
1158
1159
1160
        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
1161

1162
1163
        ptr_indptr, type_ptr_indptr, _ = _c_int_array(csr.indptr)
        ptr_data, type_ptr_data, _ = _c_float_array(csr.data)
1164

1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
        assert csr.shape[1] <= _MAX_INT32
        csr_indices = csr.indices.astype(np.int32, copy=False)

        _safe_call(_LIB.LGBM_BoosterPredictForCSR(
            self.handle,
            ptr_indptr,
            ctypes.c_int(type_ptr_indptr),
            csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
            ptr_data,
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csr.indptr)),
            ctypes.c_int64(len(csr.data)),
            ctypes.c_int64(csr.shape[1]),
            ctypes.c_int(predict_type),
            ctypes.c_int(start_iteration),
            ctypes.c_int(num_iteration),
            _c_str(self.pred_parameter),
            ctypes.byref(out_num_preds),
            preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
        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
1194
    ) -> Tuple[Union[List[scipy.sparse.csc_matrix], List[scipy.sparse.csr_matrix]], int]:
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
        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
        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)()
        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(
            self.handle,
            ptr_indptr,
            ctypes.c_int(type_ptr_indptr),
            csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
            ptr_data,
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csr.indptr)),
            ctypes.c_int64(len(csr.data)),
            ctypes.c_int64(csr.shape[1]),
            ctypes.c_int(predict_type),
            ctypes.c_int(start_iteration),
            ctypes.c_int(num_iteration),
            _c_str(self.pred_parameter),
            ctypes.c_int(matrix_type),
            out_shape.ctypes.data_as(ctypes.POINTER(ctypes.c_int64)),
            ctypes.byref(out_ptr_indptr),
            ctypes.byref(out_ptr_indices),
            ctypes.byref(out_ptr_data)))
        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

1241
1242
1243
1244
1245
1246
1247
    def __pred_for_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
1248
        """Predict for a CSR data."""
1249
        if predict_type == _C_API_PREDICT_CONTRIB:
1250
1251
1252
1253
1254
1255
            return self.__inner_predict_csr_sparse(
                csr=csr,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1256
        nrow = len(csr.indptr) - 1
1257
1258
        if nrow > _MAX_INT32:
            sections = [0] + list(np.arange(start=_MAX_INT32, stop=nrow, step=_MAX_INT32)) + [nrow]
1259
            # __get_num_preds() cannot work with nrow > MAX_INT32, so calculate overall number of predictions piecemeal
1260
            n_preds = [self.__get_num_preds(start_iteration, num_iteration, i, predict_type) for i in np.diff(sections)]
1261
            n_preds_sections = np.array([0] + n_preds, dtype=np.intp).cumsum()
1262
            preds = np.empty(sum(n_preds), dtype=np.float64)
1263
1264
            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:])):
1265
                # avoid memory consumption by arrays concatenation operations
1266
1267
1268
1269
1270
1271
1272
                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]
                )
1273
1274
            return preds, nrow
        else:
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
            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,
1285
1286
1287
1288
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
    ):
        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
        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)()
        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(
            self.handle,
            ptr_indptr,
            ctypes.c_int(type_ptr_indptr),
            csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
            ptr_data,
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csc.indptr)),
            ctypes.c_int64(len(csc.data)),
            ctypes.c_int64(csc.shape[0]),
            ctypes.c_int(predict_type),
            ctypes.c_int(start_iteration),
            ctypes.c_int(num_iteration),
            _c_str(self.pred_parameter),
            ctypes.c_int(matrix_type),
            out_shape.ctypes.data_as(ctypes.POINTER(ctypes.c_int64)),
            ctypes.byref(out_ptr_indptr),
            ctypes.byref(out_ptr_indices),
            ctypes.byref(out_ptr_data)))
        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
1335

1336
1337
1338
1339
1340
1341
1342
    def __pred_for_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        start_iteration: int,
        num_iteration: int,
        predict_type: int
    ) -> Tuple[np.ndarray, int]:
1343
        """Predict for a CSC data."""
Guolin Ke's avatar
Guolin Ke committed
1344
        nrow = csc.shape[0]
1345
        if nrow > _MAX_INT32:
1346
1347
1348
1349
1350
1351
            return self.__pred_for_csr(
                csr=csc.tocsr(),
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1352
        if predict_type == _C_API_PREDICT_CONTRIB:
1353
1354
1355
1356
1357
1358
            return self.__inner_predict_sparse_csc(
                csc=csc,
                start_iteration=start_iteration,
                num_iteration=num_iteration,
                predict_type=predict_type
            )
1359
1360
1361
1362
1363
1364
        n_preds = self.__get_num_preds(
            start_iteration=start_iteration,
            num_iteration=num_iteration,
            nrow=nrow,
            predict_type=predict_type
        )
1365
        preds = np.empty(n_preds, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
1366
1367
        out_num_preds = ctypes.c_int64(0)

1368
1369
        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
1370

1371
        assert csc.shape[0] <= _MAX_INT32
1372
        csc_indices = csc.indices.astype(np.int32, copy=False)
1373

Guolin Ke's avatar
Guolin Ke committed
1374
1375
1376
        _safe_call(_LIB.LGBM_BoosterPredictForCSC(
            self.handle,
            ptr_indptr,
1377
            ctypes.c_int(type_ptr_indptr),
1378
            csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
Guolin Ke's avatar
Guolin Ke committed
1379
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
1380
1381
1382
1383
1384
            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),
1385
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
1386
            ctypes.c_int(num_iteration),
1387
            _c_str(self.pred_parameter),
Guolin Ke's avatar
Guolin Ke committed
1388
            ctypes.byref(out_num_preds),
wxchan's avatar
wxchan committed
1389
            preds.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
wxchan's avatar
wxchan committed
1390
        if n_preds != out_num_preds.value:
1391
            raise ValueError("Wrong length for predict results")
wxchan's avatar
wxchan committed
1392
1393
        return preds, nrow

1394
    def current_iteration(self) -> int:
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
        """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(
            self.handle,
            ctypes.byref(out_cur_iter)))
        return out_cur_iter.value

wxchan's avatar
wxchan committed
1408

1409
class Dataset:
wxchan's avatar
wxchan committed
1410
    """Dataset in LightGBM."""
1411

1412
1413
    def __init__(
        self,
1414
        data: _LGBM_TrainDataType,
1415
        label: Optional[_LGBM_LabelType] = None,
1416
1417
1418
1419
        reference: Optional["Dataset"] = None,
        weight=None,
        group=None,
        init_score=None,
1420
1421
        feature_name: _LGBM_FeatureNameConfiguration = 'auto',
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = 'auto',
1422
1423
1424
        params: Optional[Dict[str, Any]] = None,
        free_raw_data: bool = True
    ):
1425
        """Initialize Dataset.
1426

wxchan's avatar
wxchan committed
1427
1428
        Parameters
        ----------
1429
        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
1430
            Data source of Dataset.
1431
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
1432
        label : list, numpy 1-D array, pandas Series / one-column DataFrame or None, optional (default=None)
1433
1434
1435
            Label of the data.
        reference : Dataset or None, optional (default=None)
            If this is Dataset for validation, training data should be used as reference.
1436
        weight : list, numpy 1-D array, pandas Series or None, optional (default=None)
1437
            Weight for each instance. Weights should be non-negative.
1438
        group : list, numpy 1-D array, pandas Series or None, optional (default=None)
1439
1440
1441
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
1442
1443
            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.
1444
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None, optional (default=None)
1445
            Init score for Dataset.
1446
        feature_name : list of str, or 'auto', optional (default="auto")
1447
1448
            Feature names.
            If 'auto' and data is pandas DataFrame, data columns names are used.
1449
        categorical_feature : list of str or int, or 'auto', optional (default="auto")
1450
1451
            Categorical features.
            If list of int, interpreted as indices.
1452
            If list of str, interpreted as feature names (need to specify ``feature_name`` as well).
1453
            If 'auto' and data is pandas DataFrame, pandas unordered categorical columns are used.
1454
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
1455
            Large values could be memory consuming. Consider using consecutive integers starting from zero.
1456
            All negative values in categorical features will be treated as missing values.
1457
            The output cannot be monotonically constrained with respect to a categorical feature.
1458
            Floating point numbers in categorical features will be rounded towards 0.
Nikita Titov's avatar
Nikita Titov committed
1459
        params : dict or None, optional (default=None)
1460
            Other parameters for Dataset.
Nikita Titov's avatar
Nikita Titov committed
1461
        free_raw_data : bool, optional (default=True)
1462
            If True, raw data is freed after constructing inner Dataset.
wxchan's avatar
wxchan committed
1463
        """
1464
        self.handle: Optional[_DatasetHandle] = None
wxchan's avatar
wxchan committed
1465
1466
1467
1468
1469
        self.data = data
        self.label = label
        self.reference = reference
        self.weight = weight
        self.group = group
1470
        self.init_score = init_score
1471
1472
        self.feature_name: _LGBM_FeatureNameConfiguration = feature_name
        self.categorical_feature: _LGBM_CategoricalFeatureConfiguration = categorical_feature
1473
        self.params = deepcopy(params)
wxchan's avatar
wxchan committed
1474
        self.free_raw_data = free_raw_data
1475
        self.used_indices: Optional[List[int]] = None
1476
        self._need_slice = True
1477
        self._predictor: Optional[_InnerPredictor] = None
1478
        self.pandas_categorical = None
1479
        self._params_back_up = None
1480
        self.version = 0
1481
        self._start_row = 0  # Used when pushing rows one by one.
wxchan's avatar
wxchan committed
1482

1483
    def __del__(self) -> None:
1484
1485
1486
1487
        try:
            self._free_handle()
        except AttributeError:
            pass
1488

1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
    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.
        """
1506
        param_str = _param_dict_to_str(self.get_params())
1507
1508
        sample_cnt = _get_sample_count(total_nrow, param_str)
        indices = np.empty(sample_cnt, dtype=np.int32)
1509
        ptr_data, _, _ = _c_int_array(indices)
1510
1511
1512
1513
        actual_sample_cnt = ctypes.c_int32(0)

        _safe_call(_LIB.LGBM_SampleIndices(
            ctypes.c_int32(total_nrow),
1514
            _c_str(param_str),
1515
1516
1517
            ptr_data,
            ctypes.byref(actual_sample_cnt),
        ))
1518
1519
        assert sample_cnt == actual_sample_cnt.value
        return indices
1520

1521
1522
1523
1524
1525
    def _init_from_ref_dataset(
        self,
        total_nrow: int,
        ref_dataset: _DatasetHandle
    ) -> 'Dataset':
1526
1527
1528
1529
1530
1531
        """Create dataset from a reference dataset.

        Parameters
        ----------
        total_nrow : int
            Number of rows expected to add to dataset.
1532
1533
        ref_dataset : object
            Handle of reference dataset to extract metadata from.
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558

        Returns
        -------
        self : Dataset
            Constructed Dataset object.
        """
        self.handle = ctypes.c_void_p()
        _safe_call(_LIB.LGBM_DatasetCreateByReference(
            ref_dataset,
            ctypes.c_int64(total_nrow),
            ctypes.byref(self.handle),
        ))
        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
        ----------
1559
        sample_data : list of numpy array
1560
            Sample data for each column.
1561
        sample_indices : list of numpy array
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
            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.
        sample_col_ptr = (ctypes.POINTER(ctypes.c_double) * ncol)()
        # c type int**
        # each int* points to start of indices for each column
        indices_col_ptr = (ctypes.POINTER(ctypes.c_int32) * ncol)()
        for i in range(ncol):
1589
1590
            sample_col_ptr[i] = _c_float_array(sample_data[i])[0]
            indices_col_ptr[i] = _c_int_array(sample_indices[i])[0]
1591
1592

        num_per_col = np.array([len(d) for d in sample_indices], dtype=np.int32)
1593
        num_per_col_ptr, _, _ = _c_int_array(num_per_col)
1594
1595

        self.handle = ctypes.c_void_p()
1596
        params_str = _param_dict_to_str(self.get_params())
1597
1598
1599
1600
1601
1602
1603
        _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),
1604
            ctypes.c_int64(total_nrow),
1605
            _c_str(params_str),
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
            ctypes.byref(self.handle),
        ))
        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)
1625
        data_ptr, data_type, _ = _c_float_array(data)
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637

        _safe_call(_LIB.LGBM_DatasetPushRows(
            self.handle,
            data_ptr,
            data_type,
            ctypes.c_int32(nrow),
            ctypes.c_int32(ncol),
            ctypes.c_int32(self._start_row),
        ))
        self._start_row += nrow
        return self

1638
    def get_params(self) -> Dict[str, Any]:
1639
1640
1641
1642
        """Get the used parameters in the Dataset.

        Returns
        -------
1643
        params : dict
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
            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",
1659
                                                "linear_tree",
1660
1661
1662
1663
                                                "max_bin",
                                                "max_bin_by_feature",
                                                "min_data_in_bin",
                                                "pre_partition",
Nikita Titov's avatar
Nikita Titov committed
1664
                                                "precise_float_parser",
1665
1666
1667
1668
1669
                                                "two_round",
                                                "use_missing",
                                                "weight_column",
                                                "zero_as_missing")
            return {k: v for k, v in self.params.items() if k in dataset_params}
1670
1671
        else:
            return {}
1672

1673
    def _free_handle(self) -> "Dataset":
1674
        if self.handle is not None:
1675
            _safe_call(_LIB.LGBM_DatasetFree(self.handle))
1676
            self.handle = None
1677
        self._need_slice = True
Guolin Ke's avatar
Guolin Ke committed
1678
1679
        if self.used_indices is not None:
            self.data = None
Nikita Titov's avatar
Nikita Titov committed
1680
        return self
wxchan's avatar
wxchan committed
1681

1682
1683
1684
1685
1686
1687
    def _set_init_score_by_predictor(
        self,
        predictor: Optional[_InnerPredictor],
        data,
        used_indices: Optional[List[int]]
    ):
Guolin Ke's avatar
Guolin Ke committed
1688
        data_has_header = False
1689
        if isinstance(data, (str, Path)):
Guolin Ke's avatar
Guolin Ke committed
1690
            # check data has header or not
1691
            data_has_header = any(self.params.get(alias, False) for alias in _ConfigAliases.get("header"))
Guolin Ke's avatar
Guolin Ke committed
1692
        num_data = self.num_data()
1693
1694
1695
        if predictor is not None:
            init_score = predictor.predict(data,
                                           raw_score=True,
1696
1697
                                           data_has_header=data_has_header)
            init_score = init_score.ravel()
1698
            if used_indices is not None:
1699
                assert not self._need_slice
1700
                if isinstance(data, (str, Path)):
1701
                    sub_init_score = np.empty(num_data * predictor.num_class, dtype=np.float64)
1702
                    assert num_data == len(used_indices)
1703
1704
                    for i in range(len(used_indices)):
                        for j in range(predictor.num_class):
1705
1706
1707
1708
                            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
1709
                new_init_score = np.empty(init_score.size, dtype=np.float64)
1710
1711
                for i in range(num_data):
                    for j in range(predictor.num_class):
1712
1713
1714
                        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:
1715
            init_score = np.zeros(self.init_score.shape, dtype=np.float64)
1716
1717
        else:
            return self
Guolin Ke's avatar
Guolin Ke committed
1718
1719
        self.set_init_score(init_score)

1720
1721
    def _lazy_init(
        self,
1722
        data: Optional[_LGBM_TrainDataType],
1723
        label: Optional[_LGBM_LabelType] = None,
1724
1725
1726
1727
1728
1729
1730
1731
1732
        reference: Optional["Dataset"] = None,
        weight=None,
        group=None,
        init_score=None,
        predictor=None,
        feature_name='auto',
        categorical_feature='auto',
        params: Optional[Dict[str, Any]] = None
    ) -> "Dataset":
wxchan's avatar
wxchan committed
1733
1734
        if data is None:
            self.handle = None
Nikita Titov's avatar
Nikita Titov committed
1735
            return self
Guolin Ke's avatar
Guolin Ke committed
1736
1737
1738
        if reference is not None:
            self.pandas_categorical = reference.pandas_categorical
            categorical_feature = reference.categorical_feature
1739
1740
1741
1742
        data, feature_name, categorical_feature, self.pandas_categorical = _data_from_pandas(data,
                                                                                             feature_name,
                                                                                             categorical_feature,
                                                                                             self.pandas_categorical)
Guolin Ke's avatar
Guolin Ke committed
1743

1744
        # process for args
wxchan's avatar
wxchan committed
1745
        params = {} if params is None else params
1746
1747
1748
        args_names = (getattr(self.__class__, '_lazy_init')
                      .__code__
                      .co_varnames[:getattr(self.__class__, '_lazy_init').__code__.co_argcount])
1749
        for key in params.keys():
1750
            if key in args_names:
1751
1752
                _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.')
1753
        # get categorical features
1754
1755
1756
1757
1758
1759
        if categorical_feature is not None:
            categorical_indices = set()
            feature_dict = {}
            if feature_name is not None:
                feature_dict = {name: i for i, name in enumerate(feature_name)}
            for name in categorical_feature:
1760
                if isinstance(name, str) and name in feature_dict:
1761
                    categorical_indices.add(feature_dict[name])
1762
                elif isinstance(name, int):
1763
1764
                    categorical_indices.add(name)
                else:
1765
                    raise TypeError(f"Wrong type({type(name).__name__}) or unknown name({name}) in categorical_feature")
1766
            if categorical_indices:
1767
1768
                for cat_alias in _ConfigAliases.get("categorical_feature"):
                    if cat_alias in params:
1769
                        # If the params[cat_alias] is equal to categorical_indices, do not report the warning.
1770
                        if not (isinstance(params[cat_alias], list) and set(params[cat_alias]) == categorical_indices):
1771
                            _log_warning(f'{cat_alias} in param dict is overridden.')
1772
                        params.pop(cat_alias, None)
1773
                params['categorical_column'] = sorted(categorical_indices)
1774

1775
        params_str = _param_dict_to_str(params)
1776
        self.params = params
1777
        # process for reference dataset
wxchan's avatar
wxchan committed
1778
        ref_dataset = None
wxchan's avatar
wxchan committed
1779
        if isinstance(reference, Dataset):
1780
            ref_dataset = reference.construct().handle
wxchan's avatar
wxchan committed
1781
1782
        elif reference is not None:
            raise TypeError('Reference dataset should be None or dataset instance')
1783
        # start construct data
1784
        if isinstance(data, (str, Path)):
wxchan's avatar
wxchan committed
1785
1786
            self.handle = ctypes.c_void_p()
            _safe_call(_LIB.LGBM_DatasetCreateFromFile(
1787
1788
                _c_str(str(data)),
                _c_str(params_str),
wxchan's avatar
wxchan committed
1789
1790
1791
1792
                ref_dataset,
                ctypes.byref(self.handle)))
        elif isinstance(data, scipy.sparse.csr_matrix):
            self.__init_from_csr(data, params_str, ref_dataset)
Guolin Ke's avatar
Guolin Ke committed
1793
1794
        elif isinstance(data, scipy.sparse.csc_matrix):
            self.__init_from_csc(data, params_str, ref_dataset)
wxchan's avatar
wxchan committed
1795
1796
        elif isinstance(data, np.ndarray):
            self.__init_from_np2d(data, params_str, ref_dataset)
1797
1798
1799
1800
1801
1802
1803
1804
1805
        elif isinstance(data, list) and len(data) > 0:
            if all(isinstance(x, np.ndarray) for x in data):
                self.__init_from_list_np2d(data, params_str, ref_dataset)
            elif all(isinstance(x, Sequence) for x in data):
                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)
1806
        elif isinstance(data, dt_DataTable):
1807
            self.__init_from_np2d(data.to_numpy(), params_str, ref_dataset)
wxchan's avatar
wxchan committed
1808
1809
1810
1811
        else:
            try:
                csr = scipy.sparse.csr_matrix(data)
                self.__init_from_csr(csr, params_str, ref_dataset)
1812
            except BaseException:
1813
                raise TypeError(f'Cannot initialize Dataset from {type(data).__name__}')
wxchan's avatar
wxchan committed
1814
1815
1816
        if label is not None:
            self.set_label(label)
        if self.get_label() is None:
1817
            raise ValueError("Label should not be None")
wxchan's avatar
wxchan committed
1818
1819
1820
1821
        if weight is not None:
            self.set_weight(weight)
        if group is not None:
            self.set_group(group)
1822
1823
        if isinstance(predictor, _InnerPredictor):
            if self._predictor is None and init_score is not None:
1824
                _log_warning("The init_score will be overridden by the prediction of init_model.")
1825
1826
1827
1828
1829
            self._set_init_score_by_predictor(
                predictor=predictor,
                data=data,
                used_indices=None
            )
1830
1831
        elif init_score is not None:
            self.set_init_score(init_score)
Guolin Ke's avatar
Guolin Ke committed
1832
        elif predictor is not None:
1833
            raise TypeError(f'Wrong predictor type {type(predictor).__name__}')
Guolin Ke's avatar
Guolin Ke committed
1834
        # set feature names
Nikita Titov's avatar
Nikita Titov committed
1835
        return self.set_feature_name(feature_name)
wxchan's avatar
wxchan committed
1836

1837
1838
    @staticmethod
    def _yield_row_from_seqlist(seqs: List[Sequence], indices: Iterable[int]):
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
        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.
1864
        sampled = np.array([row for row in self._yield_row_from_seqlist(seqs, indices)])
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
        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

1880
1881
1882
    def __init_from_seqs(
        self,
        seqs: List[Sequence],
1883
        ref_dataset: Optional[_DatasetHandle]
1884
    ) -> "Dataset":
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
        """
        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:
1899
            param_str = _param_dict_to_str(self.get_params())
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
            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

1913
1914
1915
1916
1917
1918
    def __init_from_np2d(
        self,
        mat: np.ndarray,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
1919
        """Initialize data from a 2-D numpy matrix."""
wxchan's avatar
wxchan committed
1920
1921
1922
1923
1924
1925
        if len(mat.shape) != 2:
            raise ValueError('Input numpy.ndarray must be 2 dimensional')

        self.handle = ctypes.c_void_p()
        if mat.dtype == np.float32 or mat.dtype == np.float64:
            data = np.array(mat.reshape(mat.size), dtype=mat.dtype, copy=False)
1926
        else:  # change non-float data to float data, need to copy
wxchan's avatar
wxchan committed
1927
1928
            data = np.array(mat.reshape(mat.size), dtype=np.float32)

1929
        ptr_data, type_ptr_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
1930
1931
        _safe_call(_LIB.LGBM_DatasetCreateFromMat(
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
1932
            ctypes.c_int(type_ptr_data),
1933
1934
            ctypes.c_int32(mat.shape[0]),
            ctypes.c_int32(mat.shape[1]),
1935
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
1936
            _c_str(params_str),
wxchan's avatar
wxchan committed
1937
1938
            ref_dataset,
            ctypes.byref(self.handle)))
Nikita Titov's avatar
Nikita Titov committed
1939
        return self
wxchan's avatar
wxchan committed
1940

1941
1942
1943
1944
1945
1946
    def __init_from_list_np2d(
        self,
        mats: List[np.ndarray],
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
1947
        """Initialize data from a list of 2-D numpy matrices."""
1948
        ncol = mats[0].shape[1]
1949
        nrow = np.empty((len(mats),), np.int32)
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
        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 = []
        type_ptr_data = None

        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)
1969
            else:  # change non-float data to float data, need to copy
1970
1971
                mats[i] = np.array(mat.reshape(mat.size), dtype=np.float32)

1972
            chunk_ptr_data, chunk_type_ptr_data, holder = _c_float_array(mats[i])
1973
1974
1975
1976
1977
1978
1979
1980
            if type_ptr_data is not None and chunk_type_ptr_data != type_ptr_data:
                raise ValueError('Input chunks must have same type')
            ptr_data[i] = chunk_ptr_data
            type_ptr_data = chunk_type_ptr_data
            holders.append(holder)

        self.handle = ctypes.c_void_p()
        _safe_call(_LIB.LGBM_DatasetCreateFromMats(
1981
            ctypes.c_int32(len(mats)),
1982
1983
1984
            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)),
1985
            ctypes.c_int32(ncol),
1986
            ctypes.c_int(_C_API_IS_ROW_MAJOR),
1987
            _c_str(params_str),
1988
1989
            ref_dataset,
            ctypes.byref(self.handle)))
Nikita Titov's avatar
Nikita Titov committed
1990
        return self
1991

1992
1993
1994
1995
1996
1997
    def __init_from_csr(
        self,
        csr: scipy.sparse.csr_matrix,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
1998
        """Initialize data from a CSR matrix."""
wxchan's avatar
wxchan committed
1999
        if len(csr.indices) != len(csr.data):
2000
            raise ValueError(f'Length mismatch: {len(csr.indices)} vs {len(csr.data)}')
wxchan's avatar
wxchan committed
2001
2002
        self.handle = ctypes.c_void_p()

2003
2004
        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
2005

2006
        assert csr.shape[1] <= _MAX_INT32
2007
        csr_indices = csr.indices.astype(np.int32, copy=False)
2008

wxchan's avatar
wxchan committed
2009
2010
        _safe_call(_LIB.LGBM_DatasetCreateFromCSR(
            ptr_indptr,
Guolin Ke's avatar
Guolin Ke committed
2011
            ctypes.c_int(type_ptr_indptr),
2012
            csr_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
wxchan's avatar
wxchan committed
2013
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2014
2015
2016
2017
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csr.indptr)),
            ctypes.c_int64(len(csr.data)),
            ctypes.c_int64(csr.shape[1]),
2018
            _c_str(params_str),
wxchan's avatar
wxchan committed
2019
2020
            ref_dataset,
            ctypes.byref(self.handle)))
Nikita Titov's avatar
Nikita Titov committed
2021
        return self
wxchan's avatar
wxchan committed
2022

2023
2024
2025
2026
2027
2028
    def __init_from_csc(
        self,
        csc: scipy.sparse.csc_matrix,
        params_str: str,
        ref_dataset: Optional[_DatasetHandle]
    ) -> "Dataset":
2029
        """Initialize data from a CSC matrix."""
Guolin Ke's avatar
Guolin Ke committed
2030
        if len(csc.indices) != len(csc.data):
2031
            raise ValueError(f'Length mismatch: {len(csc.indices)} vs {len(csc.data)}')
Guolin Ke's avatar
Guolin Ke committed
2032
2033
        self.handle = ctypes.c_void_p()

2034
2035
        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
2036

2037
        assert csc.shape[0] <= _MAX_INT32
2038
        csc_indices = csc.indices.astype(np.int32, copy=False)
2039

Guolin Ke's avatar
Guolin Ke committed
2040
2041
        _safe_call(_LIB.LGBM_DatasetCreateFromCSC(
            ptr_indptr,
Guolin Ke's avatar
Guolin Ke committed
2042
            ctypes.c_int(type_ptr_indptr),
2043
            csc_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
Guolin Ke's avatar
Guolin Ke committed
2044
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2045
2046
2047
2048
            ctypes.c_int(type_ptr_data),
            ctypes.c_int64(len(csc.indptr)),
            ctypes.c_int64(len(csc.data)),
            ctypes.c_int64(csc.shape[0]),
2049
            _c_str(params_str),
Guolin Ke's avatar
Guolin Ke committed
2050
2051
            ref_dataset,
            ctypes.byref(self.handle)))
Nikita Titov's avatar
Nikita Titov committed
2052
        return self
Guolin Ke's avatar
Guolin Ke committed
2053

2054
    @staticmethod
2055
2056
2057
2058
2059
2060
    def _compare_params_for_warning(
        params: Optional[Dict[str, Any]],
        other_params: Optional[Dict[str, Any]],
        ignore_keys: Set[str]
    ) -> bool:
        """Compare two dictionaries with params ignoring some keys.
2061

2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
        It is only for the warning purpose.

        Parameters
        ----------
        params : dict or None
            One dictionary with parameters to compare.
        other_params : dict or None
            Another dictionary with parameters to compare.
        ignore_keys : set
            Keys that should be ignored during comparing two dictionaries.
2072
2073
2074

        Returns
        -------
2075
2076
        compare_result : bool
          Returns whether two dictionaries with params are equal.
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
        """
        if params is None:
            params = {}
        if other_params is None:
            other_params = {}
        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

2092
    def construct(self) -> "Dataset":
2093
2094
2095
2096
2097
        """Lazy init.

        Returns
        -------
        self : Dataset
Nikita Titov's avatar
Nikita Titov committed
2098
            Constructed Dataset object.
2099
        """
2100
        if self.handle is None:
wxchan's avatar
wxchan committed
2101
            if self.reference is not None:
2102
                reference_params = self.reference.get_params()
2103
2104
                params = self.get_params()
                if params != reference_params:
2105
2106
2107
2108
2109
                    if not self._compare_params_for_warning(
                        params=params,
                        other_params=reference_params,
                        ignore_keys=_ConfigAliases.get("categorical_feature")
                    ):
2110
                        _log_warning('Overriding the parameters from Reference Dataset.')
2111
                    self._update_params(reference_params)
wxchan's avatar
wxchan committed
2112
                if self.used_indices is None:
2113
                    # create valid
2114
                    self._lazy_init(self.data, label=self.label, reference=self.reference,
2115
2116
                                    weight=self.weight, group=self.group,
                                    init_score=self.init_score, predictor=self._predictor,
2117
                                    feature_name=self.feature_name, params=self.params)
wxchan's avatar
wxchan committed
2118
                else:
2119
                    # construct subset
2120
                    used_indices = _list_to_1d_numpy(self.used_indices, np.int32, name='used_indices')
2121
                    assert used_indices.flags.c_contiguous
Guolin Ke's avatar
Guolin Ke committed
2122
                    if self.reference.group is not None:
2123
                        group_info = np.array(self.reference.group).astype(np.int32, copy=False)
2124
                        _, self.group = np.unique(np.repeat(range(len(group_info)), repeats=group_info)[self.used_indices],
2125
                                                  return_counts=True)
2126
                    self.handle = ctypes.c_void_p()
2127
                    params_str = _param_dict_to_str(self.params)
wxchan's avatar
wxchan committed
2128
                    _safe_call(_LIB.LGBM_DatasetGetSubset(
2129
                        self.reference.construct().handle,
wxchan's avatar
wxchan committed
2130
                        used_indices.ctypes.data_as(ctypes.POINTER(ctypes.c_int32)),
2131
                        ctypes.c_int32(used_indices.shape[0]),
2132
                        _c_str(params_str),
wxchan's avatar
wxchan committed
2133
                        ctypes.byref(self.handle)))
Guolin Ke's avatar
Guolin Ke committed
2134
2135
                    if not self.free_raw_data:
                        self.get_data()
Guolin Ke's avatar
Guolin Ke committed
2136
2137
                    if self.group is not None:
                        self.set_group(self.group)
wxchan's avatar
wxchan committed
2138
2139
                    if self.get_label() is None:
                        raise ValueError("Label should not be None.")
Guolin Ke's avatar
Guolin Ke committed
2140
2141
                    if isinstance(self._predictor, _InnerPredictor) and self._predictor is not self.reference._predictor:
                        self.get_data()
2142
2143
2144
2145
2146
                        self._set_init_score_by_predictor(
                            predictor=self._predictor,
                            data=self.data,
                            used_indices=used_indices
                        )
wxchan's avatar
wxchan committed
2147
            else:
2148
                # create train
2149
                self._lazy_init(self.data, label=self.label,
2150
2151
                                weight=self.weight, group=self.group,
                                init_score=self.init_score, predictor=self._predictor,
2152
                                feature_name=self.feature_name, categorical_feature=self.categorical_feature, params=self.params)
wxchan's avatar
wxchan committed
2153
2154
            if self.free_raw_data:
                self.data = None
2155
            self.feature_name = self.get_feature_name()
wxchan's avatar
wxchan committed
2156
        return self
wxchan's avatar
wxchan committed
2157

2158
2159
    def create_valid(
        self,
2160
        data: _LGBM_TrainDataType,
2161
        label: Optional[_LGBM_LabelType] = None,
2162
2163
2164
2165
2166
        weight=None,
        group=None,
        init_score=None,
        params: Optional[Dict[str, Any]] = None
    ) -> "Dataset":
2167
        """Create validation data align with current Dataset.
wxchan's avatar
wxchan committed
2168
2169
2170

        Parameters
        ----------
2171
        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
2172
            Data source of Dataset.
2173
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM) or a LightGBM Dataset binary file.
2174
        label : list, numpy 1-D array, pandas Series / one-column DataFrame or None, optional (default=None)
2175
2176
            Label of the data.
        weight : list, numpy 1-D array, pandas Series or None, optional (default=None)
2177
            Weight for each instance. Weights should be non-negative.
2178
        group : list, numpy 1-D array, pandas Series or None, optional (default=None)
2179
2180
2181
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2182
2183
            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.
2184
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None, optional (default=None)
2185
            Init score for Dataset.
Nikita Titov's avatar
Nikita Titov committed
2186
        params : dict or None, optional (default=None)
2187
            Other parameters for validation Dataset.
2188
2189
2190

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2191
2192
        valid : Dataset
            Validation Dataset with reference to self.
wxchan's avatar
wxchan committed
2193
        """
2194
        ret = Dataset(data, label=label, reference=self,
2195
                      weight=weight, group=group, init_score=init_score,
2196
                      params=params, free_raw_data=self.free_raw_data)
wxchan's avatar
wxchan committed
2197
        ret._predictor = self._predictor
2198
        ret.pandas_categorical = self.pandas_categorical
wxchan's avatar
wxchan committed
2199
        return ret
wxchan's avatar
wxchan committed
2200

2201
2202
2203
2204
2205
    def subset(
        self,
        used_indices: List[int],
        params: Optional[Dict[str, Any]] = None
    ) -> "Dataset":
2206
        """Get subset of current Dataset.
wxchan's avatar
wxchan committed
2207
2208
2209
2210

        Parameters
        ----------
        used_indices : list of int
2211
            Indices used to create the subset.
Nikita Titov's avatar
Nikita Titov committed
2212
        params : dict or None, optional (default=None)
2213
            These parameters will be passed to Dataset constructor.
2214
2215
2216
2217
2218

        Returns
        -------
        subset : Dataset
            Subset of the current Dataset.
wxchan's avatar
wxchan committed
2219
        """
wxchan's avatar
wxchan committed
2220
2221
        if params is None:
            params = self.params
wxchan's avatar
wxchan committed
2222
        ret = Dataset(None, reference=self, feature_name=self.feature_name,
2223
2224
                      categorical_feature=self.categorical_feature, params=params,
                      free_raw_data=self.free_raw_data)
wxchan's avatar
wxchan committed
2225
        ret._predictor = self._predictor
2226
        ret.pandas_categorical = self.pandas_categorical
2227
        ret.used_indices = sorted(used_indices)
wxchan's avatar
wxchan committed
2228
2229
        return ret

2230
    def save_binary(self, filename: Union[str, Path]) -> "Dataset":
2231
        """Save Dataset to a binary file.
wxchan's avatar
wxchan committed
2232

2233
2234
2235
2236
2237
        .. 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
2238
2239
        Parameters
        ----------
2240
        filename : str or pathlib.Path
wxchan's avatar
wxchan committed
2241
            Name of the output file.
Nikita Titov's avatar
Nikita Titov committed
2242
2243
2244
2245
2246

        Returns
        -------
        self : Dataset
            Returns self.
wxchan's avatar
wxchan committed
2247
2248
2249
        """
        _safe_call(_LIB.LGBM_DatasetSaveBinary(
            self.construct().handle,
2250
            _c_str(str(filename))))
Nikita Titov's avatar
Nikita Titov committed
2251
        return self
wxchan's avatar
wxchan committed
2252

2253
    def _update_params(self, params: Optional[Dict[str, Any]]) -> "Dataset":
2254
2255
        if not params:
            return self
2256
        params = deepcopy(params)
2257
2258
2259
2260
2261

        def update():
            if not self.params:
                self.params = params
            else:
2262
                self._params_back_up = deepcopy(self.params)
2263
2264
2265
2266
2267
2268
                self.params.update(params)

        if self.handle is None:
            update()
        elif params is not None:
            ret = _LIB.LGBM_DatasetUpdateParamChecking(
2269
2270
                _c_str(_param_dict_to_str(self.params)),
                _c_str(_param_dict_to_str(params)))
2271
2272
2273
2274
2275
2276
            if ret != 0:
                # could be updated if data is not freed
                if self.data is not None:
                    update()
                    self._free_handle()
                else:
2277
                    raise LightGBMError(_LIB.LGBM_GetLastError().decode('utf-8'))
Nikita Titov's avatar
Nikita Titov committed
2278
        return self
wxchan's avatar
wxchan committed
2279

2280
    def _reverse_update_params(self) -> "Dataset":
2281
        if self.handle is None:
2282
2283
            self.params = deepcopy(self._params_back_up)
            self._params_back_up = None
Nikita Titov's avatar
Nikita Titov committed
2284
        return self
2285

2286
2287
2288
2289
2290
    def set_field(
        self,
        field_name: str,
        data
    ) -> "Dataset":
wxchan's avatar
wxchan committed
2291
        """Set property into the Dataset.
wxchan's avatar
wxchan committed
2292
2293
2294

        Parameters
        ----------
2295
        field_name : str
2296
            The field name of the information.
2297
2298
        data : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None
            The data to be set.
Nikita Titov's avatar
Nikita Titov committed
2299
2300
2301
2302
2303

        Returns
        -------
        self : Dataset
            Dataset with set property.
wxchan's avatar
wxchan committed
2304
        """
2305
        if self.handle is None:
2306
            raise Exception(f"Cannot set {field_name} before construct dataset")
wxchan's avatar
wxchan committed
2307
        if data is None:
2308
            # set to None
wxchan's avatar
wxchan committed
2309
2310
            _safe_call(_LIB.LGBM_DatasetSetField(
                self.handle,
2311
                _c_str(field_name),
wxchan's avatar
wxchan committed
2312
                None,
Guolin Ke's avatar
Guolin Ke committed
2313
                ctypes.c_int(0),
2314
                ctypes.c_int(_FIELD_TYPE_MAPPER[field_name])))
Nikita Titov's avatar
Nikita Titov committed
2315
            return self
2316
        if field_name == 'init_score':
Guolin Ke's avatar
Guolin Ke committed
2317
            dtype = np.float64
2318
            if _is_1d_collection(data):
2319
                data = _list_to_1d_numpy(data, dtype, name=field_name)
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
            elif _is_2d_collection(data):
                data = _data_to_2d_numpy(data, dtype, name=field_name)
                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:
            dtype = np.int32 if field_name == 'group' else np.float32
2330
            data = _list_to_1d_numpy(data, dtype, name=field_name)
2331

2332
        if data.dtype == np.float32 or data.dtype == np.float64:
2333
            ptr_data, type_data, _ = _c_float_array(data)
wxchan's avatar
wxchan committed
2334
        elif data.dtype == np.int32:
2335
            ptr_data, type_data, _ = _c_int_array(data)
wxchan's avatar
wxchan committed
2336
        else:
2337
            raise TypeError(f"Expected np.float32/64 or np.int32, met type({data.dtype})")
2338
        if type_data != _FIELD_TYPE_MAPPER[field_name]:
2339
            raise TypeError("Input type error for set_field")
wxchan's avatar
wxchan committed
2340
2341
        _safe_call(_LIB.LGBM_DatasetSetField(
            self.handle,
2342
            _c_str(field_name),
wxchan's avatar
wxchan committed
2343
            ptr_data,
Guolin Ke's avatar
Guolin Ke committed
2344
2345
            ctypes.c_int(len(data)),
            ctypes.c_int(type_data)))
2346
        self.version += 1
Nikita Titov's avatar
Nikita Titov committed
2347
        return self
wxchan's avatar
wxchan committed
2348

2349
    def get_field(self, field_name: str) -> Optional[np.ndarray]:
wxchan's avatar
wxchan committed
2350
        """Get property from the Dataset.
wxchan's avatar
wxchan committed
2351
2352
2353

        Parameters
        ----------
2354
        field_name : str
2355
            The field name of the information.
wxchan's avatar
wxchan committed
2356
2357
2358

        Returns
        -------
2359
        info : numpy array or None
2360
            A numpy array with information from the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2361
        """
2362
        if self.handle is None:
2363
            raise Exception(f"Cannot get {field_name} before construct Dataset")
2364
2365
        tmp_out_len = ctypes.c_int(0)
        out_type = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2366
2367
2368
        ret = ctypes.POINTER(ctypes.c_void_p)()
        _safe_call(_LIB.LGBM_DatasetGetField(
            self.handle,
2369
            _c_str(field_name),
wxchan's avatar
wxchan committed
2370
2371
2372
            ctypes.byref(tmp_out_len),
            ctypes.byref(ret),
            ctypes.byref(out_type)))
2373
        if out_type.value != _FIELD_TYPE_MAPPER[field_name]:
wxchan's avatar
wxchan committed
2374
2375
2376
            raise TypeError("Return type error for get_field")
        if tmp_out_len.value == 0:
            return None
2377
        if out_type.value == _C_API_DTYPE_INT32:
2378
            arr = _cint32_array_to_numpy(ctypes.cast(ret, ctypes.POINTER(ctypes.c_int32)), tmp_out_len.value)
2379
        elif out_type.value == _C_API_DTYPE_FLOAT32:
2380
            arr = _cfloat32_array_to_numpy(ctypes.cast(ret, ctypes.POINTER(ctypes.c_float)), tmp_out_len.value)
2381
        elif out_type.value == _C_API_DTYPE_FLOAT64:
2382
            arr = _cfloat64_array_to_numpy(ctypes.cast(ret, ctypes.POINTER(ctypes.c_double)), tmp_out_len.value)
2383
        else:
wxchan's avatar
wxchan committed
2384
            raise TypeError("Unknown type")
2385
2386
2387
2388
2389
2390
        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
2391

2392
2393
    def set_categorical_feature(
        self,
2394
        categorical_feature: _LGBM_CategoricalFeatureConfiguration
2395
    ) -> "Dataset":
2396
        """Set categorical features.
2397
2398
2399

        Parameters
        ----------
2400
        categorical_feature : list of str or int, or 'auto'
2401
            Names or indices of categorical features.
Nikita Titov's avatar
Nikita Titov committed
2402
2403
2404
2405
2406

        Returns
        -------
        self : Dataset
            Dataset with set categorical features.
2407
2408
        """
        if self.categorical_feature == categorical_feature:
Nikita Titov's avatar
Nikita Titov committed
2409
            return self
2410
        if self.data is not None:
2411
2412
            if self.categorical_feature is None:
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2413
                return self._free_handle()
2414
            elif categorical_feature == 'auto':
Nikita Titov's avatar
Nikita Titov committed
2415
                return self
2416
            else:
2417
2418
2419
                if self.categorical_feature != 'auto':
                    _log_warning('categorical_feature in Dataset is overridden.\n'
                                 f'New categorical_feature is {sorted(list(categorical_feature))}')
2420
                self.categorical_feature = categorical_feature
Nikita Titov's avatar
Nikita Titov committed
2421
                return self._free_handle()
2422
        else:
2423
2424
            raise LightGBMError("Cannot set categorical feature after freed raw data, "
                                "set free_raw_data=False when construct Dataset to avoid this.")
2425

2426
2427
2428
2429
    def _set_predictor(
        self,
        predictor: Optional[_InnerPredictor]
    ) -> "Dataset":
2430
2431
2432
2433
        """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
2434
        """
2435
        if predictor is None and self._predictor is None:
Nikita Titov's avatar
Nikita Titov committed
2436
            return self
2437
2438
2439
        elif isinstance(predictor, _InnerPredictor) and isinstance(self._predictor, _InnerPredictor):
            if (predictor == self._predictor) and (predictor.current_iteration() == self._predictor.current_iteration()):
                return self
2440
        if self.handle is None:
Guolin Ke's avatar
Guolin Ke committed
2441
            self._predictor = predictor
2442
2443
        elif self.data is not None:
            self._predictor = predictor
2444
2445
2446
2447
2448
            self._set_init_score_by_predictor(
                predictor=self._predictor,
                data=self.data,
                used_indices=None
            )
2449
2450
        elif self.used_indices is not None and self.reference is not None and self.reference.data is not None:
            self._predictor = predictor
2451
2452
2453
2454
2455
            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
2456
        else:
2457
2458
            raise LightGBMError("Cannot set predictor after freed raw data, "
                                "set free_raw_data=False when construct Dataset to avoid this.")
2459
        return self
Guolin Ke's avatar
Guolin Ke committed
2460

2461
    def set_reference(self, reference: "Dataset") -> "Dataset":
2462
        """Set reference Dataset.
Guolin Ke's avatar
Guolin Ke committed
2463
2464
2465
2466

        Parameters
        ----------
        reference : Dataset
2467
            Reference that is used as a template to construct the current Dataset.
Nikita Titov's avatar
Nikita Titov committed
2468
2469
2470
2471
2472

        Returns
        -------
        self : Dataset
            Dataset with set reference.
Guolin Ke's avatar
Guolin Ke committed
2473
        """
2474
2475
2476
        self.set_categorical_feature(reference.categorical_feature) \
            .set_feature_name(reference.feature_name) \
            ._set_predictor(reference._predictor)
2477
        # we're done if self and reference share a common upstream reference
2478
        if self.get_ref_chain().intersection(reference.get_ref_chain()):
Nikita Titov's avatar
Nikita Titov committed
2479
            return self
Guolin Ke's avatar
Guolin Ke committed
2480
2481
        if self.data is not None:
            self.reference = reference
Nikita Titov's avatar
Nikita Titov committed
2482
            return self._free_handle()
Guolin Ke's avatar
Guolin Ke committed
2483
        else:
2484
2485
            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
2486

2487
    def set_feature_name(self, feature_name: Union[List[str], str]) -> "Dataset":
2488
        """Set feature name.
Guolin Ke's avatar
Guolin Ke committed
2489
2490
2491

        Parameters
        ----------
2492
        feature_name : list of str
2493
            Feature names.
Nikita Titov's avatar
Nikita Titov committed
2494
2495
2496
2497
2498

        Returns
        -------
        self : Dataset
            Dataset with set feature name.
Guolin Ke's avatar
Guolin Ke committed
2499
        """
2500
2501
        if feature_name != 'auto':
            self.feature_name = feature_name
2502
        if self.handle is not None and feature_name is not None and feature_name != 'auto':
wxchan's avatar
wxchan committed
2503
            if len(feature_name) != self.num_feature():
2504
                raise ValueError(f"Length of feature_name({len(feature_name)}) and num_feature({self.num_feature()}) don't match")
2505
            c_feature_name = [_c_str(name) for name in feature_name]
wxchan's avatar
wxchan committed
2506
2507
            _safe_call(_LIB.LGBM_DatasetSetFeatureNames(
                self.handle,
2508
                _c_array(ctypes.c_char_p, c_feature_name),
Guolin Ke's avatar
Guolin Ke committed
2509
                ctypes.c_int(len(feature_name))))
Nikita Titov's avatar
Nikita Titov committed
2510
        return self
Guolin Ke's avatar
Guolin Ke committed
2511

2512
    def set_label(self, label: Optional[_LGBM_LabelType]) -> "Dataset":
2513
        """Set label of Dataset.
Guolin Ke's avatar
Guolin Ke committed
2514
2515
2516

        Parameters
        ----------
2517
        label : list, numpy 1-D array, pandas Series / one-column DataFrame or None
2518
            The label information to be set into Dataset.
Nikita Titov's avatar
Nikita Titov committed
2519
2520
2521
2522
2523

        Returns
        -------
        self : Dataset
            Dataset with set label.
Guolin Ke's avatar
Guolin Ke committed
2524
2525
        """
        self.label = label
2526
        if self.handle is not None:
2527
2528
2529
2530
            if isinstance(label, pd_DataFrame):
                if len(label.columns) > 1:
                    raise ValueError('DataFrame for label cannot have multiple columns')
                _check_for_bad_pandas_dtypes(label.dtypes)
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
                try:
                    # most common case (no nullable dtypes)
                    label = label.to_numpy(dtype=np.float32, 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
                    label = label.astype(np.float32, copy=False).values
                except ValueError:
                    # data has nullable dtypes, but we can specify na_value argument and copy will be made
                    label = label.to_numpy(dtype=np.float32, na_value=np.nan)
                label_array = np.ravel(label)
2542
            else:
2543
                label_array = _list_to_1d_numpy(label, name='label')
2544
            self.set_field('label', label_array)
2545
            self.label = self.get_field('label')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2546
        return self
Guolin Ke's avatar
Guolin Ke committed
2547

2548
    def set_weight(self, weight) -> "Dataset":
2549
        """Set weight of each instance.
Guolin Ke's avatar
Guolin Ke committed
2550
2551
2552

        Parameters
        ----------
2553
        weight : list, numpy 1-D array, pandas Series or None
2554
            Weight to be set for each data point. Weights should be non-negative.
Nikita Titov's avatar
Nikita Titov committed
2555
2556
2557
2558
2559

        Returns
        -------
        self : Dataset
            Dataset with set weight.
Guolin Ke's avatar
Guolin Ke committed
2560
        """
2561
2562
        if weight is not None and np.all(weight == 1):
            weight = None
Guolin Ke's avatar
Guolin Ke committed
2563
        self.weight = weight
2564
        if self.handle is not None and weight is not None:
2565
            weight = _list_to_1d_numpy(weight, name='weight')
wxchan's avatar
wxchan committed
2566
            self.set_field('weight', weight)
2567
            self.weight = self.get_field('weight')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2568
        return self
Guolin Ke's avatar
Guolin Ke committed
2569

2570
    def set_init_score(self, init_score) -> "Dataset":
2571
        """Set init score of Booster to start from.
Guolin Ke's avatar
Guolin Ke committed
2572
2573
2574

        Parameters
        ----------
2575
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None
2576
            Init score for Booster.
Nikita Titov's avatar
Nikita Titov committed
2577
2578
2579
2580
2581

        Returns
        -------
        self : Dataset
            Dataset with set init score.
Guolin Ke's avatar
Guolin Ke committed
2582
2583
        """
        self.init_score = init_score
2584
        if self.handle is not None and init_score is not None:
wxchan's avatar
wxchan committed
2585
            self.set_field('init_score', init_score)
2586
            self.init_score = self.get_field('init_score')  # original values can be modified at cpp side
Nikita Titov's avatar
Nikita Titov committed
2587
        return self
Guolin Ke's avatar
Guolin Ke committed
2588

2589
    def set_group(self, group) -> "Dataset":
2590
        """Set group size of Dataset (used for ranking).
Guolin Ke's avatar
Guolin Ke committed
2591
2592
2593

        Parameters
        ----------
2594
        group : list, numpy 1-D array, pandas Series or None
2595
2596
2597
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2598
2599
            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
2600
2601
2602
2603
2604

        Returns
        -------
        self : Dataset
            Dataset with set group.
Guolin Ke's avatar
Guolin Ke committed
2605
2606
        """
        self.group = group
2607
        if self.handle is not None and group is not None:
2608
            group = _list_to_1d_numpy(group, np.int32, name='group')
wxchan's avatar
wxchan committed
2609
            self.set_field('group', group)
Nikita Titov's avatar
Nikita Titov committed
2610
        return self
Guolin Ke's avatar
Guolin Ke committed
2611

2612
    def get_feature_name(self) -> List[str]:
2613
2614
2615
2616
        """Get the names of columns (features) in the Dataset.

        Returns
        -------
2617
        feature_names : list of str
2618
2619
2620
2621
2622
2623
2624
2625
            The names of columns (features) in the Dataset.
        """
        if self.handle is None:
            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)
2626
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
2627
2628
2629
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))
        _safe_call(_LIB.LGBM_DatasetGetFeatureNames(
            self.handle,
2630
            ctypes.c_int(num_feature),
2631
            ctypes.byref(tmp_out_len),
2632
            ctypes.c_size_t(reserved_string_buffer_size),
2633
2634
2635
2636
            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")
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
        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)]
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))
            _safe_call(_LIB.LGBM_DatasetGetFeatureNames(
                self.handle,
                ctypes.c_int(num_feature),
                ctypes.byref(tmp_out_len),
                ctypes.c_size_t(actual_string_buffer_size),
                ctypes.byref(required_string_buffer_size),
                ptr_string_buffers))
2649
        return [string_buffers[i].value.decode('utf-8') for i in range(num_feature)]
2650

2651
    def get_label(self) -> Optional[np.ndarray]:
2652
        """Get the label of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2653
2654
2655

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2656
        label : numpy array or None
2657
            The label information from the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2658
        """
2659
        if self.label is None:
wxchan's avatar
wxchan committed
2660
            self.label = self.get_field('label')
Guolin Ke's avatar
Guolin Ke committed
2661
2662
        return self.label

2663
    def get_weight(self) -> Optional[np.ndarray]:
2664
        """Get the weight of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2665
2666
2667

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2668
        weight : numpy array or None
2669
            Weight for each data point from the Dataset. Weights should be non-negative.
Guolin Ke's avatar
Guolin Ke committed
2670
        """
2671
        if self.weight is None:
wxchan's avatar
wxchan committed
2672
            self.weight = self.get_field('weight')
Guolin Ke's avatar
Guolin Ke committed
2673
2674
        return self.weight

2675
    def get_init_score(self) -> Optional[np.ndarray]:
2676
        """Get the initial score of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2677
2678
2679

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2680
        init_score : numpy array or None
2681
            Init score of Booster.
Guolin Ke's avatar
Guolin Ke committed
2682
        """
2683
        if self.init_score is None:
wxchan's avatar
wxchan committed
2684
            self.init_score = self.get_field('init_score')
Guolin Ke's avatar
Guolin Ke committed
2685
2686
        return self.init_score

2687
    def get_data(self) -> Optional[_LGBM_TrainDataType]:
2688
2689
2690
2691
        """Get the raw data of the Dataset.

        Returns
        -------
2692
        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
2693
2694
2695
2696
            Raw data used in the Dataset construction.
        """
        if self.handle is None:
            raise Exception("Cannot get data before construct Dataset")
2697
        if self._need_slice and self.used_indices is not None and self.reference is not None:
Guolin Ke's avatar
Guolin Ke committed
2698
2699
2700
2701
            self.data = self.reference.data
            if self.data is not None:
                if isinstance(self.data, np.ndarray) or scipy.sparse.issparse(self.data):
                    self.data = self.data[self.used_indices, :]
2702
                elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
2703
                    self.data = self.data.iloc[self.used_indices].copy()
2704
                elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
2705
                    self.data = self.data[self.used_indices, :]
2706
2707
2708
2709
                elif isinstance(self.data, Sequence):
                    self.data = self.data[self.used_indices]
                elif isinstance(self.data, list) and len(self.data) > 0 and all(isinstance(x, Sequence) for x in self.data):
                    self.data = np.array([row for row in self._yield_row_from_seqlist(self.data, self.used_indices)])
Guolin Ke's avatar
Guolin Ke committed
2710
                else:
2711
2712
                    _log_warning(f"Cannot subset {type(self.data).__name__} type of raw data.\n"
                                 "Returning original raw data")
2713
            self._need_slice = False
Guolin Ke's avatar
Guolin Ke committed
2714
2715
2716
        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.")
2717
2718
        return self.data

Guolin Ke's avatar
Guolin Ke committed
2719
    def get_group(self):
2720
        """Get the group of the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2721
2722
2723

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
2724
        group : numpy array or None
2725
2726
2727
            Group/query data.
            Only used in the learning-to-rank task.
            sum(group) = n_samples.
2728
2729
            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.
Guolin Ke's avatar
Guolin Ke committed
2730
        """
2731
        if self.group is None:
wxchan's avatar
wxchan committed
2732
            self.group = self.get_field('group')
Guolin Ke's avatar
Guolin Ke committed
2733
2734
            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
2735
                self.group = np.diff(self.group)
Guolin Ke's avatar
Guolin Ke committed
2736
2737
        return self.group

2738
    def num_data(self) -> int:
2739
        """Get the number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2740
2741
2742

        Returns
        -------
2743
2744
        number_of_rows : int
            The number of rows in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2745
        """
2746
        if self.handle is not None:
2747
            ret = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2748
2749
2750
            _safe_call(_LIB.LGBM_DatasetGetNumData(self.handle,
                                                   ctypes.byref(ret)))
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
2751
        else:
2752
            raise LightGBMError("Cannot get num_data before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
2753

2754
    def num_feature(self) -> int:
2755
        """Get the number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2756
2757
2758

        Returns
        -------
2759
2760
        number_of_columns : int
            The number of columns (features) in the Dataset.
Guolin Ke's avatar
Guolin Ke committed
2761
        """
2762
        if self.handle is not None:
2763
            ret = ctypes.c_int(0)
wxchan's avatar
wxchan committed
2764
2765
2766
            _safe_call(_LIB.LGBM_DatasetGetNumFeature(self.handle,
                                                      ctypes.byref(ret)))
            return ret.value
Guolin Ke's avatar
Guolin Ke committed
2767
        else:
2768
            raise LightGBMError("Cannot get num_feature before construct dataset")
Guolin Ke's avatar
Guolin Ke committed
2769

2770
    def feature_num_bin(self, feature: Union[int, str]) -> int:
2771
2772
2773
2774
        """Get the number of bins for a feature.

        Parameters
        ----------
2775
2776
        feature : int or str
            Index or name of the feature.
2777
2778
2779
2780
2781
2782
2783

        Returns
        -------
        number_of_bins : int
            The number of constructed bins for the feature in the Dataset.
        """
        if self.handle is not None:
2784
            if isinstance(feature, str):
2785
2786
2787
                feature_index = self.feature_name.index(feature)
            else:
                feature_index = feature
2788
2789
            ret = ctypes.c_int(0)
            _safe_call(_LIB.LGBM_DatasetGetFeatureNumBin(self.handle,
2790
                                                         ctypes.c_int(feature_index),
2791
2792
2793
2794
2795
                                                         ctypes.byref(ret)))
            return ret.value
        else:
            raise LightGBMError("Cannot get feature_num_bin before construct dataset")

2796
    def get_ref_chain(self, ref_limit: int = 100) -> Set["Dataset"]:
2797
2798
2799
2800
2801
        """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.
2802
2803
2804
2805
2806

        Parameters
        ----------
        ref_limit : int, optional (default=100)
            The limit number of references.
2807
2808
2809

        Returns
        -------
2810
2811
2812
        ref_chain : set of Dataset
            Chain of references of the Datasets.
        """
2813
        head = self
2814
        ref_chain: Set[Dataset] = set()
2815
2816
        while len(ref_chain) < ref_limit:
            if isinstance(head, Dataset):
2817
                ref_chain.add(head)
2818
2819
2820
2821
2822
2823
                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
2824
        return ref_chain
2825

2826
    def add_features_from(self, other: "Dataset") -> "Dataset":
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
        """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.
        """
        if self.handle is None or other.handle is None:
            raise ValueError('Both source and target Datasets must be constructed before adding features')
        _safe_call(_LIB.LGBM_DatasetAddFeaturesFrom(self.handle, other.handle))
Guolin Ke's avatar
Guolin Ke committed
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
        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))
                elif scipy.sparse.issparse(other.data):
                    self.data = np.hstack((self.data, other.data.toarray()))
2854
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
2855
                    self.data = np.hstack((self.data, other.data.values))
2856
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
2857
2858
2859
2860
2861
2862
2863
                    self.data = np.hstack((self.data, other.data.to_numpy()))
                else:
                    self.data = None
            elif scipy.sparse.issparse(self.data):
                sparse_format = self.data.getformat()
                if isinstance(other.data, np.ndarray) or scipy.sparse.issparse(other.data):
                    self.data = scipy.sparse.hstack((self.data, other.data), format=sparse_format)
2864
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
2865
                    self.data = scipy.sparse.hstack((self.data, other.data.values), format=sparse_format)
2866
                elif isinstance(other.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
2867
2868
2869
                    self.data = scipy.sparse.hstack((self.data, other.data.to_numpy()), format=sparse_format)
                else:
                    self.data = None
2870
            elif isinstance(self.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
2871
2872
                if not PANDAS_INSTALLED:
                    raise LightGBMError("Cannot add features to DataFrame type of raw data "
2873
2874
                                        "without pandas installed. "
                                        "Install pandas and restart your session.")
Guolin Ke's avatar
Guolin Ke committed
2875
                if isinstance(other.data, np.ndarray):
2876
                    self.data = concat((self.data, pd_DataFrame(other.data)),
Guolin Ke's avatar
Guolin Ke committed
2877
2878
                                       axis=1, ignore_index=True)
                elif scipy.sparse.issparse(other.data):
2879
                    self.data = concat((self.data, pd_DataFrame(other.data.toarray())),
Guolin Ke's avatar
Guolin Ke committed
2880
                                       axis=1, ignore_index=True)
2881
                elif isinstance(other.data, pd_DataFrame):
Guolin Ke's avatar
Guolin Ke committed
2882
2883
                    self.data = concat((self.data, other.data),
                                       axis=1, ignore_index=True)
2884
2885
                elif isinstance(other.data, dt_DataTable):
                    self.data = concat((self.data, pd_DataFrame(other.data.to_numpy())),
Guolin Ke's avatar
Guolin Ke committed
2886
2887
2888
                                       axis=1, ignore_index=True)
                else:
                    self.data = None
2889
            elif isinstance(self.data, dt_DataTable):
Guolin Ke's avatar
Guolin Ke committed
2890
                if isinstance(other.data, np.ndarray):
2891
                    self.data = dt_DataTable(np.hstack((self.data.to_numpy(), other.data)))
Guolin Ke's avatar
Guolin Ke committed
2892
                elif scipy.sparse.issparse(other.data):
2893
2894
2895
2896
2897
                    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
2898
2899
2900
2901
2902
                else:
                    self.data = None
            else:
                self.data = None
        if self.data is None:
2903
2904
            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
2905
2906
            err_msg += ("Set free_raw_data=False when construct Dataset to avoid this"
                        if was_none else "Freeing raw data")
2907
            _log_warning(err_msg)
Guolin Ke's avatar
Guolin Ke committed
2908
        self.feature_name = self.get_feature_name()
2909
2910
        _log_warning("Reseting categorical features.\n"
                     "You can set new categorical features via ``set_categorical_feature`` method")
Guolin Ke's avatar
Guolin Ke committed
2911
2912
        self.categorical_feature = "auto"
        self.pandas_categorical = None
2913
2914
        return self

2915
    def _dump_text(self, filename: Union[str, Path]) -> "Dataset":
2916
2917
2918
2919
2920
2921
        """Save Dataset to a text file.

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

        Parameters
        ----------
2922
        filename : str or pathlib.Path
2923
2924
2925
2926
2927
2928
2929
2930
2931
            Name of the output file.

        Returns
        -------
        self : Dataset
            Returns self.
        """
        _safe_call(_LIB.LGBM_DatasetDumpText(
            self.construct().handle,
2932
            _c_str(str(filename))))
2933
2934
        return self

wxchan's avatar
wxchan committed
2935

2936
2937
2938
2939
_LGBM_CustomObjectiveFunction = Callable[
    [np.ndarray, Dataset],
    Tuple[np.ndarray, np.ndarray]
]
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
_LGBM_CustomEvalFunction = Union[
    Callable[
        [np.ndarray, Dataset],
        _LGBM_EvalFunctionResultType
    ],
    Callable[
        [np.ndarray, Dataset],
        List[_LGBM_EvalFunctionResultType]
    ]
]
2950
2951


2952
class Booster:
2953
    """Booster in LightGBM."""
2954

2955
2956
2957
2958
2959
2960
2961
    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
    ):
2962
        """Initialize the Booster.
wxchan's avatar
wxchan committed
2963
2964
2965

        Parameters
        ----------
Nikita Titov's avatar
Nikita Titov committed
2966
        params : dict or None, optional (default=None)
2967
2968
2969
            Parameters for Booster.
        train_set : Dataset or None, optional (default=None)
            Training dataset.
2970
        model_file : str, pathlib.Path or None, optional (default=None)
wxchan's avatar
wxchan committed
2971
            Path to the model file.
2972
        model_str : str or None, optional (default=None)
2973
            Model will be loaded from this string.
wxchan's avatar
wxchan committed
2974
        """
2975
        self.handle = None
2976
        self._network = False
wxchan's avatar
wxchan committed
2977
        self.__need_reload_eval_info = True
2978
        self._train_data_name = "training"
2979
        self.__set_objective_to_none = False
wxchan's avatar
wxchan committed
2980
        self.best_iteration = -1
2981
        self.best_score: _LGBM_BoosterBestScoreType = {}
2982
        params = {} if params is None else deepcopy(params)
wxchan's avatar
wxchan committed
2983
        if train_set is not None:
2984
            # Training task
wxchan's avatar
wxchan committed
2985
            if not isinstance(train_set, Dataset):
2986
                raise TypeError(f'Training data should be Dataset instance, met {type(train_set).__name__}')
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
            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"]
                )
3021
            # construct booster object
3022
3023
3024
            train_set.construct()
            # copy the parameters from train_set
            params.update(train_set.get_params())
3025
            params_str = _param_dict_to_str(params)
3026
            self.handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
3027
            _safe_call(_LIB.LGBM_BoosterCreate(
3028
                train_set.handle,
3029
                _c_str(params_str),
wxchan's avatar
wxchan committed
3030
                ctypes.byref(self.handle)))
3031
            # save reference to data
wxchan's avatar
wxchan committed
3032
            self.train_set = train_set
3033
3034
            self.valid_sets: List[Dataset] = []
            self.name_valid_sets: List[str] = []
wxchan's avatar
wxchan committed
3035
            self.__num_dataset = 1
Guolin Ke's avatar
Guolin Ke committed
3036
3037
            self.__init_predictor = train_set._predictor
            if self.__init_predictor is not None:
wxchan's avatar
wxchan committed
3038
3039
                _safe_call(_LIB.LGBM_BoosterMerge(
                    self.handle,
Guolin Ke's avatar
Guolin Ke committed
3040
                    self.__init_predictor.handle))
Guolin Ke's avatar
Guolin Ke committed
3041
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3042
3043
3044
3045
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
                self.handle,
                ctypes.byref(out_num_class)))
            self.__num_class = out_num_class.value
3046
            # buffer for inner predict
wxchan's avatar
wxchan committed
3047
3048
3049
            self.__inner_predict_buffer = [None]
            self.__is_predicted_cur_iter = [False]
            self.__get_eval_info()
3050
            self.pandas_categorical = train_set.pandas_categorical
3051
            self.train_set_version = train_set.version
wxchan's avatar
wxchan committed
3052
        elif model_file is not None:
3053
            # Prediction task
Guolin Ke's avatar
Guolin Ke committed
3054
            out_num_iterations = ctypes.c_int(0)
3055
            self.handle = ctypes.c_void_p()
wxchan's avatar
wxchan committed
3056
            _safe_call(_LIB.LGBM_BoosterCreateFromModelfile(
3057
                _c_str(str(model_file)),
wxchan's avatar
wxchan committed
3058
3059
                ctypes.byref(out_num_iterations),
                ctypes.byref(self.handle)))
Guolin Ke's avatar
Guolin Ke committed
3060
            out_num_class = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3061
3062
3063
3064
            _safe_call(_LIB.LGBM_BoosterGetNumClasses(
                self.handle,
                ctypes.byref(out_num_class)))
            self.__num_class = out_num_class.value
3065
            self.pandas_categorical = _load_pandas_categorical(file_name=model_file)
3066
3067
3068
            if params:
                _log_warning('Ignoring params argument, using parameters from model file.')
            params = self._get_loaded_param()
3069
        elif model_str is not None:
3070
            self.model_from_string(model_str)
wxchan's avatar
wxchan committed
3071
        else:
3072
3073
            raise TypeError('Need at least one training dataset or model file or model string '
                            'to create Booster instance')
3074
        self.params = params
wxchan's avatar
wxchan committed
3075

3076
    def __del__(self) -> None:
3077
        try:
3078
            if self._network:
3079
3080
3081
3082
3083
3084
3085
3086
                self.free_network()
        except AttributeError:
            pass
        try:
            if self.handle is not None:
                _safe_call(_LIB.LGBM_BoosterFree(self.handle))
        except AttributeError:
            pass
wxchan's avatar
wxchan committed
3087

3088
    def __copy__(self) -> "Booster":
wxchan's avatar
wxchan committed
3089
3090
        return self.__deepcopy__(None)

3091
    def __deepcopy__(self, _) -> "Booster":
3092
        model_str = self.model_to_string(num_iteration=-1)
3093
        booster = Booster(model_str=model_str)
3094
        return booster
wxchan's avatar
wxchan committed
3095

3096
    def __getstate__(self) -> Dict[str, Any]:
wxchan's avatar
wxchan committed
3097
3098
3099
3100
3101
        this = self.__dict__.copy()
        handle = this['handle']
        this.pop('train_set', None)
        this.pop('valid_sets', None)
        if handle is not None:
3102
            this["handle"] = self.model_to_string(num_iteration=-1)
wxchan's avatar
wxchan committed
3103
3104
        return this

3105
    def __setstate__(self, state: Dict[str, Any]) -> None:
3106
3107
        model_str = state.get('handle', None)
        if model_str is not None:
wxchan's avatar
wxchan committed
3108
            handle = ctypes.c_void_p()
Guolin Ke's avatar
Guolin Ke committed
3109
            out_num_iterations = ctypes.c_int(0)
3110
            _safe_call(_LIB.LGBM_BoosterLoadModelFromString(
3111
                _c_str(model_str),
3112
3113
                ctypes.byref(out_num_iterations),
                ctypes.byref(handle)))
wxchan's avatar
wxchan committed
3114
3115
3116
            state['handle'] = handle
        self.__dict__.update(state)

3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
    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)
        ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
        _safe_call(_LIB.LGBM_BoosterGetLoadedParam(
            self.handle,
            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)
            ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
            _safe_call(_LIB.LGBM_BoosterGetLoadedParam(
                self.handle,
                ctypes.c_int64(actual_len),
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
        return json.loads(string_buffer.value.decode('utf-8'))

3139
    def free_dataset(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3140
3141
3142
3143
3144
3145
3146
        """Free Booster's Datasets.

        Returns
        -------
        self : Booster
            Booster without Datasets.
        """
wxchan's avatar
wxchan committed
3147
3148
        self.__dict__.pop('train_set', None)
        self.__dict__.pop('valid_sets', None)
3149
        self.__num_dataset = 0
Nikita Titov's avatar
Nikita Titov committed
3150
        return self
wxchan's avatar
wxchan committed
3151

3152
    def _free_buffer(self) -> "Booster":
3153
3154
        self.__inner_predict_buffer = []
        self.__is_predicted_cur_iter = []
Nikita Titov's avatar
Nikita Titov committed
3155
        return self
3156

3157
3158
3159
3160
3161
3162
3163
    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":
3164
3165
3166
3167
        """Set the network configuration.

        Parameters
        ----------
3168
        machines : list, set or str
3169
            Names of machines.
Nikita Titov's avatar
Nikita Titov committed
3170
        local_listen_port : int, optional (default=12400)
3171
            TCP listen port for local machines.
Nikita Titov's avatar
Nikita Titov committed
3172
        listen_time_out : int, optional (default=120)
3173
            Socket time-out in minutes.
Nikita Titov's avatar
Nikita Titov committed
3174
        num_machines : int, optional (default=1)
3175
            The number of machines for distributed learning application.
Nikita Titov's avatar
Nikita Titov committed
3176
3177
3178
3179
3180

        Returns
        -------
        self : Booster
            Booster with set network.
3181
        """
3182
3183
        if isinstance(machines, (list, set)):
            machines = ','.join(machines)
3184
        _safe_call(_LIB.LGBM_NetworkInit(_c_str(machines),
3185
3186
3187
                                         ctypes.c_int(local_listen_port),
                                         ctypes.c_int(listen_time_out),
                                         ctypes.c_int(num_machines)))
3188
        self._network = True
Nikita Titov's avatar
Nikita Titov committed
3189
        return self
3190

3191
    def free_network(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3192
3193
3194
3195
3196
3197
3198
        """Free Booster's network.

        Returns
        -------
        self : Booster
            Booster with freed network.
        """
3199
        _safe_call(_LIB.LGBM_NetworkFree())
3200
        self._network = False
Nikita Titov's avatar
Nikita Titov committed
3201
        return self
3202

3203
    def trees_to_dataframe(self) -> pd_DataFrame:
3204
3205
        """Parse the fitted model and return in an easy-to-read pandas DataFrame.

3206
3207
3208
3209
        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.
3210
3211
3212
3213
3214
            - ``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.
3215
3216
            - ``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.
3217
            - ``decision_type`` : str, logical operator describing how to compare a value to ``threshold``.
3218
3219
              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.
3220
3221
            - ``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.
3222
            - ``value`` : float64, predicted value for this leaf node, multiplied by the learning rate.
3223
            - ``weight`` : float64 or int64, sum of Hessian (second-order derivative of objective), summed over observations that fall in this node.
3224
3225
            - ``count`` : int64, number of records in the training data that fall into this node.

3226
3227
3228
3229
3230
3231
        Returns
        -------
        result : pandas DataFrame
            Returns a pandas DataFrame of the parsed model.
        """
        if not PANDAS_INSTALLED:
3232
3233
            raise LightGBMError('This method cannot be run without pandas installed. '
                                'You must install pandas and restart your session to use this method.')
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244

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

        def _is_split_node(tree):
            return 'split_index' in tree.keys()

        def create_node_record(tree, node_depth=1, tree_index=None,
                               feature_names=None, parent_node=None):

            def _get_node_index(tree, tree_index):
3245
                tree_num = f'{tree_index}-' if tree_index is not None else ''
3246
3247
3248
                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
3249
3250
                node_num = tree.get('split_index' if is_split else 'leaf_index', 0)
                return f"{tree_num}{node_type}{node_num}"
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262

            def _get_split_feature(tree, feature_names):
                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

            def _is_single_node_tree(tree):
3263
                return set(tree.keys()) == {'leaf_value'}
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336

            # Create the node record, and populate universal data members
            node = OrderedDict()
            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

        def tree_dict_to_node_list(tree, node_depth=1, tree_index=None,
                                   feature_names=None, parent_node=None):

            node = create_node_record(tree,
                                      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(
                        tree[child],
                        node_depth=node_depth + 1,
                        tree_index=tree_index,
                        feature_names=feature_names,
                        parent_node=node['node_index'])
                    # 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']:
            model_list.extend(tree_dict_to_node_list(tree['tree_structure'],
                                                     tree_index=tree['tree_index'],
                                                     feature_names=feature_names))

3337
        return pd_DataFrame(model_list, columns=model_list[0].keys())
3338

3339
    def set_train_data_name(self, name: str) -> "Booster":
3340
3341
3342
3343
        """Set the name to the training Dataset.

        Parameters
        ----------
3344
        name : str
Nikita Titov's avatar
Nikita Titov committed
3345
3346
3347
3348
3349
3350
            Name for the training Dataset.

        Returns
        -------
        self : Booster
            Booster with set training Dataset name.
3351
        """
3352
        self._train_data_name = name
Nikita Titov's avatar
Nikita Titov committed
3353
        return self
wxchan's avatar
wxchan committed
3354

3355
    def add_valid(self, data: Dataset, name: str) -> "Booster":
3356
        """Add validation data.
wxchan's avatar
wxchan committed
3357
3358
3359
3360

        Parameters
        ----------
        data : Dataset
3361
            Validation data.
3362
        name : str
3363
            Name of validation data.
Nikita Titov's avatar
Nikita Titov committed
3364
3365
3366
3367
3368

        Returns
        -------
        self : Booster
            Booster with set validation data.
wxchan's avatar
wxchan committed
3369
        """
Guolin Ke's avatar
Guolin Ke committed
3370
        if not isinstance(data, Dataset):
3371
            raise TypeError(f'Validation data should be Dataset instance, met {type(data).__name__}')
Guolin Ke's avatar
Guolin Ke committed
3372
        if data._predictor is not self.__init_predictor:
3373
3374
            raise LightGBMError("Add validation data failed, "
                                "you should use same predictor for these data")
wxchan's avatar
wxchan committed
3375
3376
        _safe_call(_LIB.LGBM_BoosterAddValidData(
            self.handle,
wxchan's avatar
wxchan committed
3377
            data.construct().handle))
wxchan's avatar
wxchan committed
3378
3379
3380
3381
3382
        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
3383
        return self
wxchan's avatar
wxchan committed
3384

3385
    def reset_parameter(self, params: Dict[str, Any]) -> "Booster":
3386
        """Reset parameters of Booster.
wxchan's avatar
wxchan committed
3387
3388
3389
3390

        Parameters
        ----------
        params : dict
3391
            New parameters for Booster.
Nikita Titov's avatar
Nikita Titov committed
3392
3393
3394
3395
3396

        Returns
        -------
        self : Booster
            Booster with new parameters.
wxchan's avatar
wxchan committed
3397
        """
3398
        params_str = _param_dict_to_str(params)
wxchan's avatar
wxchan committed
3399
3400
3401
        if params_str:
            _safe_call(_LIB.LGBM_BoosterResetParameter(
                self.handle,
3402
                _c_str(params_str)))
Guolin Ke's avatar
Guolin Ke committed
3403
        self.params.update(params)
Nikita Titov's avatar
Nikita Titov committed
3404
        return self
wxchan's avatar
wxchan committed
3405

3406
3407
3408
3409
3410
    def update(
        self,
        train_set: Optional[Dataset] = None,
        fobj: Optional[_LGBM_CustomObjectiveFunction] = None
    ) -> bool:
Nikita Titov's avatar
Nikita Titov committed
3411
        """Update Booster for one iteration.
3412

wxchan's avatar
wxchan committed
3413
3414
        Parameters
        ----------
3415
3416
3417
3418
        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
3419
            Customized objective function.
3420
3421
3422
            Should accept two parameters: preds, train_data,
            and return (grad, hess).

3423
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
3424
                    The predicted values.
3425
3426
                    Predicted values are returned before any transformation,
                    e.g. they are raw margin instead of probability of positive class for binary task.
3427
3428
                train_data : Dataset
                    The training dataset.
3429
                grad : numpy 1-D array or numpy 2-D array (for multi-class task)
3430
3431
                    The value of the first order derivative (gradient) of the loss
                    with respect to the elements of preds for each sample point.
3432
                hess : numpy 1-D array or numpy 2-D array (for multi-class task)
3433
3434
                    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
3435

3436
            For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes],
3437
            and grad and hess should be returned in the same format.
3438

wxchan's avatar
wxchan committed
3439
3440
        Returns
        -------
3441
3442
        is_finished : bool
            Whether the update was successfully finished.
wxchan's avatar
wxchan committed
3443
        """
3444
        # need reset training data
3445
3446
3447
3448
3449
3450
        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
3451
            if not isinstance(train_set, Dataset):
3452
                raise TypeError(f'Training data should be Dataset instance, met {type(train_set).__name__}')
Guolin Ke's avatar
Guolin Ke committed
3453
            if train_set._predictor is not self.__init_predictor:
3454
3455
                raise LightGBMError("Replace training data failed, "
                                    "you should use same predictor for these data")
wxchan's avatar
wxchan committed
3456
3457
3458
            self.train_set = train_set
            _safe_call(_LIB.LGBM_BoosterResetTrainingData(
                self.handle,
wxchan's avatar
wxchan committed
3459
                self.train_set.construct().handle))
wxchan's avatar
wxchan committed
3460
            self.__inner_predict_buffer[0] = None
3461
            self.train_set_version = self.train_set.version
wxchan's avatar
wxchan committed
3462
3463
        is_finished = ctypes.c_int(0)
        if fobj is None:
3464
            if self.__set_objective_to_none:
3465
                raise LightGBMError('Cannot update due to null objective function.')
wxchan's avatar
wxchan committed
3466
3467
3468
            _safe_call(_LIB.LGBM_BoosterUpdateOneIter(
                self.handle,
                ctypes.byref(is_finished)))
3469
            self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
3470
3471
            return is_finished.value == 1
        else:
3472
            if not self.__set_objective_to_none:
Nikita Titov's avatar
Nikita Titov committed
3473
                self.reset_parameter({"objective": "none"}).__set_objective_to_none = True
wxchan's avatar
wxchan committed
3474
3475
3476
            grad, hess = fobj(self.__inner_predict(0), self.train_set)
            return self.__boost(grad, hess)

3477
3478
3479
3480
3481
    def __boost(
        self,
        grad: np.ndarray,
        hess: np.ndarray
    ) -> bool:
3482
        """Boost Booster for one iteration with customized gradient statistics.
Nikita Titov's avatar
Nikita Titov committed
3483

Nikita Titov's avatar
Nikita Titov committed
3484
3485
        .. note::

3486
3487
            Score is returned before any transformation,
            e.g. it is raw margin instead of probability of positive class for binary task.
3488
            For multi-class task, score are numpy 2-D array of shape = [n_samples, n_classes],
3489
            and grad and hess should be returned in the same format.
3490

wxchan's avatar
wxchan committed
3491
3492
        Parameters
        ----------
3493
        grad : numpy 1-D array or numpy 2-D array (for multi-class task)
3494
3495
            The value of the first order derivative (gradient) of the loss
            with respect to the elements of score for each sample point.
3496
        hess : numpy 1-D array or numpy 2-D array (for multi-class task)
3497
3498
            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
3499
3500
3501

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3502
3503
        is_finished : bool
            Whether the boost was successfully finished.
wxchan's avatar
wxchan committed
3504
        """
3505
3506
3507
        if self.__num_class > 1:
            grad = grad.ravel(order='F')
            hess = hess.ravel(order='F')
3508
3509
        grad = _list_to_1d_numpy(grad, name='gradient')
        hess = _list_to_1d_numpy(hess, name='hessian')
3510
3511
        assert grad.flags.c_contiguous
        assert hess.flags.c_contiguous
wxchan's avatar
wxchan committed
3512
        if len(grad) != len(hess):
3513
3514
            raise ValueError(f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) don't match")
        num_train_data = self.train_set.num_data()
3515
        if len(grad) != num_train_data * self.__num_class:
3516
3517
3518
            raise ValueError(
                f"Lengths of gradient ({len(grad)}) and Hessian ({len(hess)}) "
                f"don't match training data length ({num_train_data}) * "
3519
                f"number of models per one iteration ({self.__num_class})"
3520
            )
wxchan's avatar
wxchan committed
3521
3522
3523
3524
3525
3526
        is_finished = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterUpdateOneIterCustom(
            self.handle,
            grad.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
            hess.ctypes.data_as(ctypes.POINTER(ctypes.c_float)),
            ctypes.byref(is_finished)))
3527
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
wxchan's avatar
wxchan committed
3528
3529
        return is_finished.value == 1

3530
    def rollback_one_iter(self) -> "Booster":
Nikita Titov's avatar
Nikita Titov committed
3531
3532
3533
3534
3535
3536
3537
        """Rollback one iteration.

        Returns
        -------
        self : Booster
            Booster with rolled back one iteration.
        """
wxchan's avatar
wxchan committed
3538
3539
        _safe_call(_LIB.LGBM_BoosterRollbackOneIter(
            self.handle))
3540
        self.__is_predicted_cur_iter = [False for _ in range(self.__num_dataset)]
Nikita Titov's avatar
Nikita Titov committed
3541
        return self
wxchan's avatar
wxchan committed
3542

3543
    def current_iteration(self) -> int:
3544
3545
3546
3547
3548
3549
3550
        """Get the index of the current iteration.

        Returns
        -------
        cur_iter : int
            The index of the current iteration.
        """
Guolin Ke's avatar
Guolin Ke committed
3551
        out_cur_iter = ctypes.c_int(0)
wxchan's avatar
wxchan committed
3552
3553
3554
3555
3556
        _safe_call(_LIB.LGBM_BoosterGetCurrentIteration(
            self.handle,
            ctypes.byref(out_cur_iter)))
        return out_cur_iter.value

3557
    def num_model_per_iteration(self) -> int:
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
        """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(
            self.handle,
            ctypes.byref(model_per_iter)))
        return model_per_iter.value

3571
    def num_trees(self) -> int:
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
        """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(
            self.handle,
            ctypes.byref(num_trees)))
        return num_trees.value

3585
    def upper_bound(self) -> float:
3586
3587
3588
3589
        """Get upper bound value of a model.

        Returns
        -------
3590
        upper_bound : float
3591
3592
3593
3594
3595
3596
3597
3598
            Upper bound value of the model.
        """
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetUpperBoundValue(
            self.handle,
            ctypes.byref(ret)))
        return ret.value

3599
    def lower_bound(self) -> float:
3600
3601
3602
3603
        """Get lower bound value of a model.

        Returns
        -------
3604
        lower_bound : float
3605
3606
3607
3608
3609
3610
3611
3612
            Lower bound value of the model.
        """
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetLowerBoundValue(
            self.handle,
            ctypes.byref(ret)))
        return ret.value

3613
3614
3615
3616
3617
3618
    def eval(
        self,
        data: Dataset,
        name: str,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
3619
        """Evaluate for data.
wxchan's avatar
wxchan committed
3620
3621
3622

        Parameters
        ----------
3623
3624
        data : Dataset
            Data for the evaluating.
3625
        name : str
3626
            Name of the data.
3627
        feval : callable, list of callable, or None, optional (default=None)
3628
            Customized evaluation function.
3629
            Each evaluation function should accept two parameters: preds, eval_data,
3630
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
3631

3632
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
3633
                    The predicted values.
3634
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
3635
                    If custom objective function is used, predicted values are returned before any transformation,
3636
                    e.g. they are raw margin instead of probability of positive class for binary task in this case.
3637
                eval_data : Dataset
3638
                    A ``Dataset`` to evaluate.
3639
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
3640
                    The name of evaluation function (without whitespace).
3641
3642
3643
3644
3645
                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
3646
3647
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3648
        result : list
3649
            List with (dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
3650
        """
Guolin Ke's avatar
Guolin Ke committed
3651
3652
        if not isinstance(data, Dataset):
            raise TypeError("Can only eval for Dataset instance")
wxchan's avatar
wxchan committed
3653
3654
3655
3656
        data_idx = -1
        if data is self.train_set:
            data_idx = 0
        else:
3657
            for i in range(len(self.valid_sets)):
wxchan's avatar
wxchan committed
3658
3659
3660
                if data is self.valid_sets[i]:
                    data_idx = i + 1
                    break
3661
        # need to push new valid data
wxchan's avatar
wxchan committed
3662
3663
3664
3665
3666
3667
        if data_idx == -1:
            self.add_valid(data, name)
            data_idx = self.__num_dataset - 1

        return self.__inner_eval(name, data_idx, feval)

3668
3669
3670
3671
    def eval_train(
        self,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
3672
        """Evaluate for training data.
wxchan's avatar
wxchan committed
3673
3674
3675

        Parameters
        ----------
3676
        feval : callable, list of callable, or None, optional (default=None)
3677
            Customized evaluation function.
3678
            Each evaluation function should accept two parameters: preds, eval_data,
3679
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
3680

3681
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
3682
                    The predicted values.
3683
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
3684
                    If custom objective function is used, predicted values are returned before any transformation,
3685
                    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
3686
                eval_data : Dataset
3687
                    The training dataset.
3688
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
3689
                    The name of evaluation function (without whitespace).
3690
3691
3692
3693
3694
                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
3695
3696
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3697
        result : list
3698
            List with (train_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
3699
        """
3700
        return self.__inner_eval(self._train_data_name, 0, feval)
wxchan's avatar
wxchan committed
3701

3702
3703
3704
3705
    def eval_valid(
        self,
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]] = None
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
3706
        """Evaluate for validation data.
wxchan's avatar
wxchan committed
3707
3708
3709

        Parameters
        ----------
3710
        feval : callable, list of callable, or None, optional (default=None)
3711
            Customized evaluation function.
3712
            Each evaluation function should accept two parameters: preds, eval_data,
3713
            and return (eval_name, eval_result, is_higher_better) or list of such tuples.
3714

3715
                preds : numpy 1-D array or numpy 2-D array (for multi-class task)
3716
                    The predicted values.
3717
                    For multi-class task, preds are numpy 2-D array of shape = [n_samples, n_classes].
3718
                    If custom objective function is used, predicted values are returned before any transformation,
3719
                    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
3720
                eval_data : Dataset
3721
                    The validation dataset.
3722
                eval_name : str
Andrew Ziem's avatar
Andrew Ziem committed
3723
                    The name of evaluation function (without whitespace).
3724
3725
3726
3727
3728
                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
3729
3730
        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3731
        result : list
3732
            List with (validation_dataset_name, eval_name, eval_result, is_higher_better) tuples.
wxchan's avatar
wxchan committed
3733
        """
3734
        return [item for i in range(1, self.__num_dataset)
wxchan's avatar
wxchan committed
3735
                for item in self.__inner_eval(self.name_valid_sets[i - 1], i, feval)]
wxchan's avatar
wxchan committed
3736

3737
3738
3739
3740
3741
3742
3743
    def save_model(
        self,
        filename: Union[str, Path],
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
        importance_type: str = 'split'
    ) -> "Booster":
3744
        """Save Booster to file.
wxchan's avatar
wxchan committed
3745
3746
3747

        Parameters
        ----------
3748
        filename : str or pathlib.Path
3749
            Filename to save Booster.
3750
3751
3752
3753
        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
3754
        start_iteration : int, optional (default=0)
3755
            Start index of the iteration that should be saved.
3756
        importance_type : str, optional (default="split")
3757
3758
3759
            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
3760
3761
3762
3763
3764

        Returns
        -------
        self : Booster
            Returns self.
wxchan's avatar
wxchan committed
3765
        """
3766
        if num_iteration is None:
3767
            num_iteration = self.best_iteration
3768
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
3769
3770
        _safe_call(_LIB.LGBM_BoosterSaveModel(
            self.handle,
3771
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
3772
            ctypes.c_int(num_iteration),
3773
            ctypes.c_int(importance_type_int),
3774
            _c_str(str(filename))))
3775
        _dump_pandas_categorical(self.pandas_categorical, filename)
Nikita Titov's avatar
Nikita Titov committed
3776
        return self
wxchan's avatar
wxchan committed
3777

3778
3779
3780
3781
3782
    def shuffle_models(
        self,
        start_iteration: int = 0,
        end_iteration: int = -1
    ) -> "Booster":
3783
        """Shuffle models.
Nikita Titov's avatar
Nikita Titov committed
3784

3785
3786
3787
        Parameters
        ----------
        start_iteration : int, optional (default=0)
3788
            The first iteration that will be shuffled.
3789
3790
        end_iteration : int, optional (default=-1)
            The last iteration that will be shuffled.
3791
            If <= 0, means the last available iteration.
3792

Nikita Titov's avatar
Nikita Titov committed
3793
3794
3795
3796
        Returns
        -------
        self : Booster
            Booster with shuffled models.
3797
        """
3798
3799
        _safe_call(_LIB.LGBM_BoosterShuffleModels(
            self.handle,
Guolin Ke's avatar
Guolin Ke committed
3800
3801
            ctypes.c_int(start_iteration),
            ctypes.c_int(end_iteration)))
Nikita Titov's avatar
Nikita Titov committed
3802
        return self
3803

3804
    def model_from_string(self, model_str: str) -> "Booster":
3805
3806
3807
3808
        """Load Booster from a string.

        Parameters
        ----------
3809
        model_str : str
3810
3811
3812
3813
            Model will be loaded from this string.

        Returns
        -------
Nikita Titov's avatar
Nikita Titov committed
3814
        self : Booster
3815
3816
            Loaded Booster object.
        """
3817
3818
3819
3820
        if self.handle is not None:
            _safe_call(_LIB.LGBM_BoosterFree(self.handle))
        self._free_buffer()
        self.handle = ctypes.c_void_p()
3821
3822
        out_num_iterations = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterLoadModelFromString(
3823
            _c_str(model_str),
3824
3825
3826
3827
3828
3829
3830
            ctypes.byref(out_num_iterations),
            ctypes.byref(self.handle)))
        out_num_class = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterGetNumClasses(
            self.handle,
            ctypes.byref(out_num_class)))
        self.__num_class = out_num_class.value
3831
        self.pandas_categorical = _load_pandas_categorical(model_str=model_str)
3832
3833
        return self

3834
3835
3836
3837
3838
3839
    def model_to_string(
        self,
        num_iteration: Optional[int] = None,
        start_iteration: int = 0,
        importance_type: str = 'split'
    ) -> str:
3840
        """Save Booster to string.
3841

3842
3843
3844
3845
3846
3847
        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
3848
        start_iteration : int, optional (default=0)
3849
            Start index of the iteration that should be saved.
3850
        importance_type : str, optional (default="split")
3851
3852
3853
            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.
3854
3855
3856

        Returns
        -------
3857
        str_repr : str
3858
3859
            String representation of Booster.
        """
3860
        if num_iteration is None:
3861
            num_iteration = self.best_iteration
3862
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
3863
        buffer_len = 1 << 20
3864
        tmp_out_len = ctypes.c_int64(0)
3865
3866
3867
3868
        string_buffer = ctypes.create_string_buffer(buffer_len)
        ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
        _safe_call(_LIB.LGBM_BoosterSaveModelToString(
            self.handle,
3869
            ctypes.c_int(start_iteration),
3870
            ctypes.c_int(num_iteration),
3871
            ctypes.c_int(importance_type_int),
3872
            ctypes.c_int64(buffer_len),
3873
3874
3875
            ctypes.byref(tmp_out_len),
            ptr_string_buffer))
        actual_len = tmp_out_len.value
3876
        # if buffer length is not long enough, re-allocate a buffer
3877
3878
3879
3880
3881
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
            ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
            _safe_call(_LIB.LGBM_BoosterSaveModelToString(
                self.handle,
3882
                ctypes.c_int(start_iteration),
3883
                ctypes.c_int(num_iteration),
3884
                ctypes.c_int(importance_type_int),
3885
                ctypes.c_int64(actual_len),
3886
3887
                ctypes.byref(tmp_out_len),
                ptr_string_buffer))
3888
        ret = string_buffer.value.decode('utf-8')
3889
3890
        ret += _dump_pandas_categorical(self.pandas_categorical)
        return ret
3891

3892
3893
3894
3895
3896
3897
3898
    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
3899
        """Dump Booster to JSON format.
wxchan's avatar
wxchan committed
3900

3901
3902
        Parameters
        ----------
3903
3904
3905
3906
        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
3907
        start_iteration : int, optional (default=0)
3908
            Start index of the iteration that should be dumped.
3909
        importance_type : str, optional (default="split")
3910
3911
3912
            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.
3913
3914
3915
3916
3917
3918
3919
3920
3921
        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.
3922

wxchan's avatar
wxchan committed
3923
3924
        Returns
        -------
3925
        json_repr : dict
Nikita Titov's avatar
Nikita Titov committed
3926
            JSON format of Booster.
wxchan's avatar
wxchan committed
3927
        """
3928
        if num_iteration is None:
3929
            num_iteration = self.best_iteration
3930
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
wxchan's avatar
wxchan committed
3931
        buffer_len = 1 << 20
3932
        tmp_out_len = ctypes.c_int64(0)
wxchan's avatar
wxchan committed
3933
3934
3935
3936
        string_buffer = ctypes.create_string_buffer(buffer_len)
        ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
        _safe_call(_LIB.LGBM_BoosterDumpModel(
            self.handle,
3937
            ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
3938
            ctypes.c_int(num_iteration),
3939
            ctypes.c_int(importance_type_int),
3940
            ctypes.c_int64(buffer_len),
wxchan's avatar
wxchan committed
3941
            ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
3942
            ptr_string_buffer))
wxchan's avatar
wxchan committed
3943
        actual_len = tmp_out_len.value
3944
        # if buffer length is not long enough, reallocate a buffer
wxchan's avatar
wxchan committed
3945
3946
3947
3948
3949
        if actual_len > buffer_len:
            string_buffer = ctypes.create_string_buffer(actual_len)
            ptr_string_buffer = ctypes.c_char_p(*[ctypes.addressof(string_buffer)])
            _safe_call(_LIB.LGBM_BoosterDumpModel(
                self.handle,
3950
                ctypes.c_int(start_iteration),
Guolin Ke's avatar
Guolin Ke committed
3951
                ctypes.c_int(num_iteration),
3952
                ctypes.c_int(importance_type_int),
3953
                ctypes.c_int64(actual_len),
wxchan's avatar
wxchan committed
3954
                ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
3955
                ptr_string_buffer))
3956
        ret = json.loads(string_buffer.value.decode('utf-8'), object_hook=object_hook)
3957
        ret['pandas_categorical'] = json.loads(json.dumps(self.pandas_categorical,
3958
                                                          default=_json_default_with_numpy))
3959
        return ret
wxchan's avatar
wxchan committed
3960

3961
3962
    def predict(
        self,
3963
        data: _LGBM_PredictDataType,
3964
3965
3966
3967
3968
3969
3970
3971
        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
3972
    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
3973
        """Make a prediction.
wxchan's avatar
wxchan committed
3974
3975
3976

        Parameters
        ----------
3977
        data : str, pathlib.Path, numpy array, pandas DataFrame, H2O DataTable's Frame or scipy.sparse
3978
            Data source for prediction.
3979
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
3980
        start_iteration : int, optional (default=0)
3981
            Start index of the iteration to predict.
3982
            If <= 0, starts from the first iteration.
3983
        num_iteration : int or None, optional (default=None)
3984
3985
3986
3987
            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).
3988
3989
3990
3991
        raw_score : bool, optional (default=False)
            Whether to predict raw scores.
        pred_leaf : bool, optional (default=False)
            Whether to predict leaf index.
3992
3993
        pred_contrib : bool, optional (default=False)
            Whether to predict feature contributions.
3994

Nikita Titov's avatar
Nikita Titov committed
3995
3996
3997
3998
3999
4000
4001
            .. 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.
4002

4003
4004
        data_has_header : bool, optional (default=False)
            Whether the data has header.
4005
            Used only if data is str.
4006
4007
4008
        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.
4009
4010
        **kwargs
            Other parameters for the prediction.
wxchan's avatar
wxchan committed
4011
4012
4013

        Returns
        -------
4014
        result : numpy array, scipy.sparse or list of scipy.sparse
4015
            Prediction result.
4016
            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
4017
        """
4018
        predictor = self._to_predictor(deepcopy(kwargs))
4019
        if num_iteration is None:
4020
            if start_iteration <= 0:
4021
4022
4023
4024
                num_iteration = self.best_iteration
            else:
                num_iteration = -1
        return predictor.predict(data, start_iteration, num_iteration,
4025
                                 raw_score, pred_leaf, pred_contrib,
4026
                                 data_has_header, validate_features)
wxchan's avatar
wxchan committed
4027

4028
4029
    def refit(
        self,
4030
        data: _LGBM_TrainDataType,
4031
        label,
4032
4033
        decay_rate: float = 0.9,
        reference: Optional[Dataset] = None,
4034
4035
4036
        weight=None,
        group=None,
        init_score=None,
4037
4038
        feature_name: _LGBM_FeatureNameConfiguration = 'auto',
        categorical_feature: _LGBM_CategoricalFeatureConfiguration = 'auto',
4039
4040
4041
        dataset_params: Optional[Dict[str, Any]] = None,
        free_raw_data: bool = True,
        validate_features: bool = False,
4042
        **kwargs
4043
    ) -> "Booster":
Guolin Ke's avatar
Guolin Ke committed
4044
4045
4046
4047
        """Refit the existing Booster by new data.

        Parameters
        ----------
4048
        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
4049
            Data source for refit.
4050
            If str or pathlib.Path, it represents the path to a text file (CSV, TSV, or LibSVM).
4051
        label : list, numpy 1-D array or pandas Series / one-column DataFrame
Guolin Ke's avatar
Guolin Ke committed
4052
4053
            Label for refit.
        decay_rate : float, optional (default=0.9)
4054
4055
            Decay rate of refit,
            will use ``leaf_output = decay_rate * old_leaf_output + (1.0 - decay_rate) * new_leaf_output`` to refit trees.
4056
4057
4058
        reference : Dataset or None, optional (default=None)
            Reference for ``data``.
        weight : list, numpy 1-D array, pandas Series or None, optional (default=None)
4059
            Weight for each ``data`` instance. Weights should be non-negative.
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
        group : list, numpy 1-D array, pandas Series or None, optional (default=None)
            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.
        init_score : list, list of lists (for multi-class task), numpy array, pandas Series, pandas DataFrame (for multi-class task), or None, optional (default=None)
            Init score for ``data``.
        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.
        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.
4076
            All values in categorical features will be cast to int32 and thus should be less than int32 max value (2147483647).
4077
4078
4079
            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.
4080
            Floating point numbers in categorical features will be rounded towards 0.
4081
4082
4083
4084
        dataset_params : dict or None, optional (default=None)
            Other parameters for Dataset ``data``.
        free_raw_data : bool, optional (default=True)
            If True, raw data is freed after constructing inner Dataset for ``data``.
4085
4086
4087
        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.
4088
4089
        **kwargs
            Other parameters for refit.
4090
            These parameters will be passed to ``predict`` method.
Guolin Ke's avatar
Guolin Ke committed
4091
4092
4093
4094
4095
4096

        Returns
        -------
        result : Booster
            Refitted Booster.
        """
4097
4098
        if self.__set_objective_to_none:
            raise LightGBMError('Cannot refit due to null objective function.')
4099
4100
        if dataset_params is None:
            dataset_params = {}
4101
        predictor = self._to_predictor(deepcopy(kwargs))
4102
        leaf_preds = predictor.predict(data, -1, pred_leaf=True, validate_features=validate_features)
4103
        nrow, ncol = leaf_preds.shape
4104
        out_is_linear = ctypes.c_int(0)
4105
4106
4107
        _safe_call(_LIB.LGBM_BoosterGetLinear(
            self.handle,
            ctypes.byref(out_is_linear)))
Nikita Titov's avatar
Nikita Titov committed
4108
4109
4110
4111
4112
        new_params = _choose_param_value(
            main_param_name="linear_tree",
            params=self.params,
            default_value=None
        )
4113
        new_params["linear_tree"] = bool(out_is_linear.value)
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
        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,
        )
4127
        new_params['refit_decay_rate'] = decay_rate
4128
        new_booster = Booster(new_params, train_set)
Guolin Ke's avatar
Guolin Ke committed
4129
4130
4131
4132
4133
        # Copy models
        _safe_call(_LIB.LGBM_BoosterMerge(
            new_booster.handle,
            predictor.handle))
        leaf_preds = leaf_preds.reshape(-1)
4134
        ptr_data, _, _ = _c_int_array(leaf_preds)
Guolin Ke's avatar
Guolin Ke committed
4135
4136
4137
        _safe_call(_LIB.LGBM_BoosterRefit(
            new_booster.handle,
            ptr_data,
4138
4139
            ctypes.c_int32(nrow),
            ctypes.c_int32(ncol)))
4140
        new_booster._network = self._network
Guolin Ke's avatar
Guolin Ke committed
4141
4142
        return new_booster

4143
    def get_leaf_output(self, tree_id: int, leaf_id: int) -> float:
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
        """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.
        """
4158
4159
4160
4161
4162
4163
4164
4165
        ret = ctypes.c_double(0)
        _safe_call(_LIB.LGBM_BoosterGetLeafValue(
            self.handle,
            ctypes.c_int(tree_id),
            ctypes.c_int(leaf_id),
            ctypes.byref(ret)))
        return ret.value

4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
    def set_leaf_output(
        self,
        tree_id: int,
        leaf_id: int,
        value: float,
    ) -> 'Booster':
        """Set 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.
        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(
                self.handle,
                ctypes.c_int(tree_id),
                ctypes.c_int(leaf_id),
                ctypes.c_double(value)
            )
        )
        return self

4198
4199
4200
4201
    def _to_predictor(
        self,
        pred_parameter: Optional[Dict[str, Any]] = None
    ) -> _InnerPredictor:
4202
        """Convert to predictor."""
4203
        predictor = _InnerPredictor(booster_handle=self.handle, pred_parameter=pred_parameter)
4204
        predictor.pandas_categorical = self.pandas_categorical
wxchan's avatar
wxchan committed
4205
4206
        return predictor

4207
    def num_feature(self) -> int:
4208
4209
4210
4211
4212
4213
4214
        """Get number of features.

        Returns
        -------
        num_feature : int
            The number of features.
        """
4215
4216
4217
4218
4219
4220
        out_num_feature = ctypes.c_int(0)
        _safe_call(_LIB.LGBM_BoosterGetNumFeature(
            self.handle,
            ctypes.byref(out_num_feature)))
        return out_num_feature.value

4221
    def feature_name(self) -> List[str]:
4222
        """Get names of features.
wxchan's avatar
wxchan committed
4223
4224
4225

        Returns
        -------
4226
        result : list of str
4227
            List with names of features.
wxchan's avatar
wxchan committed
4228
        """
4229
        num_feature = self.num_feature()
4230
        # Get name of features
wxchan's avatar
wxchan committed
4231
        tmp_out_len = ctypes.c_int(0)
4232
4233
        reserved_string_buffer_size = 255
        required_string_buffer_size = ctypes.c_size_t(0)
4234
        string_buffers = [ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(num_feature)]
wxchan's avatar
wxchan committed
4235
4236
4237
        ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))
        _safe_call(_LIB.LGBM_BoosterGetFeatureNames(
            self.handle,
4238
            ctypes.c_int(num_feature),
wxchan's avatar
wxchan committed
4239
            ctypes.byref(tmp_out_len),
4240
            ctypes.c_size_t(reserved_string_buffer_size),
4241
            ctypes.byref(required_string_buffer_size),
wxchan's avatar
wxchan committed
4242
4243
4244
            ptr_string_buffers))
        if num_feature != tmp_out_len.value:
            raise ValueError("Length of feature names doesn't equal with num_feature")
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
        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)]
            ptr_string_buffers = (ctypes.c_char_p * num_feature)(*map(ctypes.addressof, string_buffers))
            _safe_call(_LIB.LGBM_BoosterGetFeatureNames(
                self.handle,
                ctypes.c_int(num_feature),
                ctypes.byref(tmp_out_len),
                ctypes.c_size_t(actual_string_buffer_size),
                ctypes.byref(required_string_buffer_size),
                ptr_string_buffers))
4257
        return [string_buffers[i].value.decode('utf-8') for i in range(num_feature)]
wxchan's avatar
wxchan committed
4258

4259
4260
4261
4262
4263
    def feature_importance(
        self,
        importance_type: str = 'split',
        iteration: Optional[int] = None
    ) -> np.ndarray:
4264
        """Get feature importances.
4265

4266
4267
        Parameters
        ----------
4268
        importance_type : str, optional (default="split")
4269
4270
4271
            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.
4272
4273
4274
4275
        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).
4276

4277
4278
        Returns
        -------
4279
4280
        result : numpy array
            Array with feature importances.
4281
        """
4282
4283
        if iteration is None:
            iteration = self.best_iteration
4284
        importance_type_int = _FEATURE_IMPORTANCE_TYPE_MAPPER[importance_type]
4285
        result = np.empty(self.num_feature(), dtype=np.float64)
4286
4287
4288
4289
4290
        _safe_call(_LIB.LGBM_BoosterFeatureImportance(
            self.handle,
            ctypes.c_int(iteration),
            ctypes.c_int(importance_type_int),
            result.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
4291
        if importance_type_int == _C_API_FEATURE_IMPORTANCE_SPLIT:
4292
            return result.astype(np.int32)
4293
4294
        else:
            return result
4295

4296
4297
4298
4299
4300
4301
    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]:
4302
4303
4304
4305
        """Get split value histogram for the specified feature.

        Parameters
        ----------
4306
        feature : int or str
4307
4308
            The feature name or index the histogram is calculated for.
            If int, interpreted as index.
4309
            If str, interpreted as name.
4310

Nikita Titov's avatar
Nikita Titov committed
4311
4312
4313
            .. warning::

                Categorical features are not supported.
4314

4315
        bins : int, str or None, optional (default=None)
4316
4317
4318
            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.
4319
            If str, it should be one from the list of the supported values by ``numpy.histogram()`` function.
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
        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.
        """
4334
        def add(root: Dict[str, Any]) -> None:
4335
4336
            """Recursively add thresholds."""
            if 'split_index' in root:  # non-leaf
4337
                if feature_names is not None and isinstance(feature, str):
4338
4339
4340
4341
                    split_feature = feature_names[root['split_feature']]
                else:
                    split_feature = root['split_feature']
                if split_feature == feature:
4342
                    if isinstance(root['threshold'], str):
4343
4344
4345
                        raise LightGBMError('Cannot compute split value histogram for the categorical feature')
                    else:
                        values.append(root['threshold'])
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
                add(root['left_child'])
                add(root['right_child'])

        model = self.dump_model()
        feature_names = model.get('feature_names')
        tree_infos = model['tree_info']
        values = []
        for tree_info in tree_infos:
            add(tree_info['tree_structure'])

4356
        if bins is None or isinstance(bins, int) and xgboost_style:
4357
4358
4359
4360
4361
4362
4363
            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:
4364
                return pd_DataFrame(ret, columns=['SplitValue', 'Count'])
4365
4366
4367
4368
4369
            else:
                return ret
        else:
            return hist, bin_edges

4370
4371
4372
4373
    def __inner_eval(
        self,
        data_name: str,
        data_idx: int,
4374
        feval: Optional[Union[_LGBM_CustomEvalFunction, List[_LGBM_CustomEvalFunction]]]
4375
    ) -> List[_LGBM_BoosterEvalMethodResultType]:
4376
        """Evaluate training or validation data."""
wxchan's avatar
wxchan committed
4377
        if data_idx >= self.__num_dataset:
4378
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
4379
4380
4381
        self.__get_eval_info()
        ret = []
        if self.__num_inner_eval > 0:
4382
            result = np.empty(self.__num_inner_eval, dtype=np.float64)
Guolin Ke's avatar
Guolin Ke committed
4383
            tmp_out_len = ctypes.c_int(0)
wxchan's avatar
wxchan committed
4384
4385
            _safe_call(_LIB.LGBM_BoosterGetEval(
                self.handle,
Guolin Ke's avatar
Guolin Ke committed
4386
                ctypes.c_int(data_idx),
wxchan's avatar
wxchan committed
4387
                ctypes.byref(tmp_out_len),
Guolin Ke's avatar
Guolin Ke committed
4388
                result.ctypes.data_as(ctypes.POINTER(ctypes.c_double))))
wxchan's avatar
wxchan committed
4389
            if tmp_out_len.value != self.__num_inner_eval:
4390
                raise ValueError("Wrong length of eval results")
4391
            for i in range(self.__num_inner_eval):
4392
4393
                ret.append((data_name, self.__name_inner_eval[i],
                            result[i], self.__higher_better_inner_eval[i]))
4394
4395
        if callable(feval):
            feval = [feval]
wxchan's avatar
wxchan committed
4396
4397
4398
4399
4400
        if feval is not None:
            if data_idx == 0:
                cur_data = self.train_set
            else:
                cur_data = self.valid_sets[data_idx - 1]
4401
4402
4403
4404
4405
4406
4407
4408
4409
            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
4410
4411
4412
                    ret.append((data_name, eval_name, val, is_higher_better))
        return ret

4413
    def __inner_predict(self, data_idx: int) -> np.ndarray:
4414
        """Predict for training and validation dataset."""
wxchan's avatar
wxchan committed
4415
        if data_idx >= self.__num_dataset:
4416
            raise ValueError("Data_idx should be smaller than number of dataset")
wxchan's avatar
wxchan committed
4417
4418
4419
4420
4421
        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
4422
            self.__inner_predict_buffer[data_idx] = np.empty(n_preds, dtype=np.float64)
4423
        # avoid to predict many time in one iteration
wxchan's avatar
wxchan committed
4424
4425
        if not self.__is_predicted_cur_iter[data_idx]:
            tmp_out_len = ctypes.c_int64(0)
Guolin Ke's avatar
Guolin Ke committed
4426
            data_ptr = self.__inner_predict_buffer[data_idx].ctypes.data_as(ctypes.POINTER(ctypes.c_double))
wxchan's avatar
wxchan committed
4427
4428
            _safe_call(_LIB.LGBM_BoosterGetPredict(
                self.handle,
Guolin Ke's avatar
Guolin Ke committed
4429
                ctypes.c_int(data_idx),
wxchan's avatar
wxchan committed
4430
4431
4432
                ctypes.byref(tmp_out_len),
                data_ptr))
            if tmp_out_len.value != len(self.__inner_predict_buffer[data_idx]):
4433
                raise ValueError(f"Wrong length of predict results for data {data_idx}")
wxchan's avatar
wxchan committed
4434
            self.__is_predicted_cur_iter[data_idx] = True
4435
4436
4437
4438
4439
        result = self.__inner_predict_buffer[data_idx]
        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
4440

4441
    def __get_eval_info(self) -> None:
4442
        """Get inner evaluation count and names."""
wxchan's avatar
wxchan committed
4443
4444
        if self.__need_reload_eval_info:
            self.__need_reload_eval_info = False
Guolin Ke's avatar
Guolin Ke committed
4445
            out_num_eval = ctypes.c_int(0)
4446
            # Get num of inner evals
wxchan's avatar
wxchan committed
4447
4448
4449
4450
4451
            _safe_call(_LIB.LGBM_BoosterGetEvalCounts(
                self.handle,
                ctypes.byref(out_num_eval)))
            self.__num_inner_eval = out_num_eval.value
            if self.__num_inner_eval > 0:
4452
                # Get name of eval metrics
Guolin Ke's avatar
Guolin Ke committed
4453
                tmp_out_len = ctypes.c_int(0)
4454
4455
4456
                reserved_string_buffer_size = 255
                required_string_buffer_size = ctypes.c_size_t(0)
                string_buffers = [
4457
                    ctypes.create_string_buffer(reserved_string_buffer_size) for _ in range(self.__num_inner_eval)
4458
                ]
wxchan's avatar
wxchan committed
4459
                ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))
wxchan's avatar
wxchan committed
4460
4461
                _safe_call(_LIB.LGBM_BoosterGetEvalNames(
                    self.handle,
4462
                    ctypes.c_int(self.__num_inner_eval),
wxchan's avatar
wxchan committed
4463
                    ctypes.byref(tmp_out_len),
4464
                    ctypes.c_size_t(reserved_string_buffer_size),
4465
                    ctypes.byref(required_string_buffer_size),
wxchan's avatar
wxchan committed
4466
4467
                    ptr_string_buffers))
                if self.__num_inner_eval != tmp_out_len.value:
4468
                    raise ValueError("Length of eval names doesn't equal with num_evals")
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
                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)
                    ]
                    ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))
                    _safe_call(_LIB.LGBM_BoosterGetEvalNames(
                        self.handle,
                        ctypes.c_int(self.__num_inner_eval),
                        ctypes.byref(tmp_out_len),
                        ctypes.c_size_t(actual_string_buffer_size),
                        ctypes.byref(required_string_buffer_size),
                        ptr_string_buffers))
                self.__name_inner_eval = [
                    string_buffers[i].value.decode('utf-8') for i in range(self.__num_inner_eval)
                ]
                self.__higher_better_inner_eval = [
                    name.startswith(('auc', 'ndcg@', 'map@', 'average_precision')) for name in self.__name_inner_eval
                ]