File size: 18,837 Bytes
7885a28 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 |
import inspect
from collections import defaultdict
from functools import partial
import numpy as np
from numpy.testing import assert_array_equal
from sklearn.base import (
BaseEstimator,
ClassifierMixin,
MetaEstimatorMixin,
RegressorMixin,
TransformerMixin,
clone,
)
from sklearn.metrics._scorer import _Scorer, mean_squared_error
from sklearn.model_selection import BaseCrossValidator
from sklearn.model_selection._split import GroupsConsumerMixin
from sklearn.utils._metadata_requests import (
SIMPLE_METHODS,
)
from sklearn.utils.metadata_routing import (
MetadataRouter,
MethodMapping,
process_routing,
)
from sklearn.utils.multiclass import _check_partial_fit_first_call
def record_metadata(obj, record_default=True, **kwargs):
"""Utility function to store passed metadata to a method of obj.
If record_default is False, kwargs whose values are "default" are skipped.
This is so that checks on keyword arguments whose default was not changed
are skipped.
"""
stack = inspect.stack()
callee = stack[1].function
caller = stack[2].function
if not hasattr(obj, "_records"):
obj._records = defaultdict(lambda: defaultdict(list))
if not record_default:
kwargs = {
key: val
for key, val in kwargs.items()
if not isinstance(val, str) or (val != "default")
}
obj._records[callee][caller].append(kwargs)
def check_recorded_metadata(obj, method, parent, split_params=tuple(), **kwargs):
"""Check whether the expected metadata is passed to the object's method.
Parameters
----------
obj : estimator object
sub-estimator to check routed params for
method : str
sub-estimator's method where metadata is routed to, or otherwise in
the context of metadata routing referred to as 'callee'
parent : str
the parent method which should have called `method`, or otherwise in
the context of metadata routing referred to as 'caller'
split_params : tuple, default=empty
specifies any parameters which are to be checked as being a subset
of the original values
**kwargs : dict
passed metadata
"""
all_records = (
getattr(obj, "_records", dict()).get(method, dict()).get(parent, list())
)
for record in all_records:
# first check that the names of the metadata passed are the same as
# expected. The names are stored as keys in `record`.
assert set(kwargs.keys()) == set(
record.keys()
), f"Expected {kwargs.keys()} vs {record.keys()}"
for key, value in kwargs.items():
recorded_value = record[key]
# The following condition is used to check for any specified parameters
# being a subset of the original values
if key in split_params and recorded_value is not None:
assert np.isin(recorded_value, value).all()
else:
if isinstance(recorded_value, np.ndarray):
assert_array_equal(recorded_value, value)
else:
assert (
recorded_value is value
), f"Expected {recorded_value} vs {value}. Method: {method}"
record_metadata_not_default = partial(record_metadata, record_default=False)
def assert_request_is_empty(metadata_request, exclude=None):
"""Check if a metadata request dict is empty.
One can exclude a method or a list of methods from the check using the
``exclude`` parameter. If metadata_request is a MetadataRouter, then
``exclude`` can be of the form ``{"object" : [method, ...]}``.
"""
if isinstance(metadata_request, MetadataRouter):
for name, route_mapping in metadata_request:
if exclude is not None and name in exclude:
_exclude = exclude[name]
else:
_exclude = None
assert_request_is_empty(route_mapping.router, exclude=_exclude)
return
exclude = [] if exclude is None else exclude
for method in SIMPLE_METHODS:
if method in exclude:
continue
mmr = getattr(metadata_request, method)
props = [
prop
for prop, alias in mmr.requests.items()
if isinstance(alias, str) or alias is not None
]
assert not props
def assert_request_equal(request, dictionary):
for method, requests in dictionary.items():
mmr = getattr(request, method)
assert mmr.requests == requests
empty_methods = [method for method in SIMPLE_METHODS if method not in dictionary]
for method in empty_methods:
assert not len(getattr(request, method).requests)
class _Registry(list):
# This list is used to get a reference to the sub-estimators, which are not
# necessarily stored on the metaestimator. We need to override __deepcopy__
# because the sub-estimators are probably cloned, which would result in a
# new copy of the list, but we need copy and deep copy both to return the
# same instance.
def __deepcopy__(self, memo):
return self
def __copy__(self):
return self
class ConsumingRegressor(RegressorMixin, BaseEstimator):
"""A regressor consuming metadata.
Parameters
----------
registry : list, default=None
If a list, the estimator will append itself to the list in order to have
a reference to the estimator later on. Since that reference is not
required in all tests, registration can be skipped by leaving this value
as None.
"""
def __init__(self, registry=None):
self.registry = registry
def partial_fit(self, X, y, sample_weight="default", metadata="default"):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return self
def fit(self, X, y, sample_weight="default", metadata="default"):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return self
def predict(self, X, y=None, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return np.zeros(shape=(len(X),))
def score(self, X, y, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return 1
class NonConsumingClassifier(ClassifierMixin, BaseEstimator):
"""A classifier which accepts no metadata on any method."""
def __init__(self, alpha=0.0):
self.alpha = alpha
def fit(self, X, y):
self.classes_ = np.unique(y)
self.coef_ = np.ones_like(X)
return self
def partial_fit(self, X, y, classes=None):
return self
def decision_function(self, X):
return self.predict(X)
def predict(self, X):
y_pred = np.empty(shape=(len(X),))
y_pred[: len(X) // 2] = 0
y_pred[len(X) // 2 :] = 1
return y_pred
def predict_proba(self, X):
# dummy probabilities to support predict_proba
y_proba = np.empty(shape=(len(X), 2))
y_proba[: len(X) // 2, :] = np.asarray([1.0, 0.0])
y_proba[len(X) // 2 :, :] = np.asarray([0.0, 1.0])
return y_proba
def predict_log_proba(self, X):
# dummy probabilities to support predict_log_proba
return self.predict_proba(X)
class NonConsumingRegressor(RegressorMixin, BaseEstimator):
"""A classifier which accepts no metadata on any method."""
def fit(self, X, y):
return self
def partial_fit(self, X, y):
return self
def predict(self, X):
return np.ones(len(X)) # pragma: no cover
class ConsumingClassifier(ClassifierMixin, BaseEstimator):
"""A classifier consuming metadata.
Parameters
----------
registry : list, default=None
If a list, the estimator will append itself to the list in order to have
a reference to the estimator later on. Since that reference is not
required in all tests, registration can be skipped by leaving this value
as None.
alpha : float, default=0
This parameter is only used to test the ``*SearchCV`` objects, and
doesn't do anything.
"""
def __init__(self, registry=None, alpha=0.0):
self.alpha = alpha
self.registry = registry
def partial_fit(
self, X, y, classes=None, sample_weight="default", metadata="default"
):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
_check_partial_fit_first_call(self, classes)
return self
def fit(self, X, y, sample_weight="default", metadata="default"):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
self.classes_ = np.unique(y)
self.coef_ = np.ones_like(X)
return self
def predict(self, X, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
y_score = np.empty(shape=(len(X),), dtype="int8")
y_score[len(X) // 2 :] = 0
y_score[: len(X) // 2] = 1
return y_score
def predict_proba(self, X, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
y_proba = np.empty(shape=(len(X), 2))
y_proba[: len(X) // 2, :] = np.asarray([1.0, 0.0])
y_proba[len(X) // 2 :, :] = np.asarray([0.0, 1.0])
return y_proba
def predict_log_proba(self, X, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return np.zeros(shape=(len(X), 2))
def decision_function(self, X, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
y_score = np.empty(shape=(len(X),))
y_score[len(X) // 2 :] = 0
y_score[: len(X) // 2] = 1
return y_score
def score(self, X, y, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return 1
class ConsumingTransformer(TransformerMixin, BaseEstimator):
"""A transformer which accepts metadata on fit and transform.
Parameters
----------
registry : list, default=None
If a list, the estimator will append itself to the list in order to have
a reference to the estimator later on. Since that reference is not
required in all tests, registration can be skipped by leaving this value
as None.
"""
def __init__(self, registry=None):
self.registry = registry
def fit(self, X, y=None, sample_weight="default", metadata="default"):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
self.fitted_ = True
return self
def transform(self, X, sample_weight="default", metadata="default"):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return X + 1
def fit_transform(self, X, y, sample_weight="default", metadata="default"):
# implementing ``fit_transform`` is necessary since
# ``TransformerMixin.fit_transform`` doesn't route any metadata to
# ``transform``, while here we want ``transform`` to receive
# ``sample_weight`` and ``metadata``.
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return self.fit(X, y, sample_weight=sample_weight, metadata=metadata).transform(
X, sample_weight=sample_weight, metadata=metadata
)
def inverse_transform(self, X, sample_weight=None, metadata=None):
record_metadata_not_default(
self, sample_weight=sample_weight, metadata=metadata
)
return X - 1
class ConsumingNoFitTransformTransformer(BaseEstimator):
"""A metadata consuming transformer that doesn't inherit from
TransformerMixin, and thus doesn't implement `fit_transform`. Note that
TransformerMixin's `fit_transform` doesn't route metadata to `transform`."""
def __init__(self, registry=None):
self.registry = registry
def fit(self, X, y=None, sample_weight=None, metadata=None):
if self.registry is not None:
self.registry.append(self)
record_metadata(self, sample_weight=sample_weight, metadata=metadata)
return self
def transform(self, X, sample_weight=None, metadata=None):
record_metadata(self, sample_weight=sample_weight, metadata=metadata)
return X
class ConsumingScorer(_Scorer):
def __init__(self, registry=None):
super().__init__(
score_func=mean_squared_error, sign=1, kwargs={}, response_method="predict"
)
self.registry = registry
def _score(self, method_caller, clf, X, y, **kwargs):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(self, **kwargs)
sample_weight = kwargs.get("sample_weight", None)
return super()._score(method_caller, clf, X, y, sample_weight=sample_weight)
class ConsumingSplitter(GroupsConsumerMixin, BaseCrossValidator):
def __init__(self, registry=None):
self.registry = registry
def split(self, X, y=None, groups="default", metadata="default"):
if self.registry is not None:
self.registry.append(self)
record_metadata_not_default(self, groups=groups, metadata=metadata)
split_index = len(X) // 2
train_indices = list(range(0, split_index))
test_indices = list(range(split_index, len(X)))
yield test_indices, train_indices
yield train_indices, test_indices
def get_n_splits(self, X=None, y=None, groups=None, metadata=None):
return 2
def _iter_test_indices(self, X=None, y=None, groups=None):
split_index = len(X) // 2
train_indices = list(range(0, split_index))
test_indices = list(range(split_index, len(X)))
yield test_indices
yield train_indices
class MetaRegressor(MetaEstimatorMixin, RegressorMixin, BaseEstimator):
"""A meta-regressor which is only a router."""
def __init__(self, estimator):
self.estimator = estimator
def fit(self, X, y, **fit_params):
params = process_routing(self, "fit", **fit_params)
self.estimator_ = clone(self.estimator).fit(X, y, **params.estimator.fit)
def get_metadata_routing(self):
router = MetadataRouter(owner=self.__class__.__name__).add(
estimator=self.estimator,
method_mapping=MethodMapping().add(caller="fit", callee="fit"),
)
return router
class WeightedMetaRegressor(MetaEstimatorMixin, RegressorMixin, BaseEstimator):
"""A meta-regressor which is also a consumer."""
def __init__(self, estimator, registry=None):
self.estimator = estimator
self.registry = registry
def fit(self, X, y, sample_weight=None, **fit_params):
if self.registry is not None:
self.registry.append(self)
record_metadata(self, sample_weight=sample_weight)
params = process_routing(self, "fit", sample_weight=sample_weight, **fit_params)
self.estimator_ = clone(self.estimator).fit(X, y, **params.estimator.fit)
return self
def predict(self, X, **predict_params):
params = process_routing(self, "predict", **predict_params)
return self.estimator_.predict(X, **params.estimator.predict)
def get_metadata_routing(self):
router = (
MetadataRouter(owner=self.__class__.__name__)
.add_self_request(self)
.add(
estimator=self.estimator,
method_mapping=MethodMapping()
.add(caller="fit", callee="fit")
.add(caller="predict", callee="predict"),
)
)
return router
class WeightedMetaClassifier(MetaEstimatorMixin, ClassifierMixin, BaseEstimator):
"""A meta-estimator which also consumes sample_weight itself in ``fit``."""
def __init__(self, estimator, registry=None):
self.estimator = estimator
self.registry = registry
def fit(self, X, y, sample_weight=None, **kwargs):
if self.registry is not None:
self.registry.append(self)
record_metadata(self, sample_weight=sample_weight)
params = process_routing(self, "fit", sample_weight=sample_weight, **kwargs)
self.estimator_ = clone(self.estimator).fit(X, y, **params.estimator.fit)
return self
def get_metadata_routing(self):
router = (
MetadataRouter(owner=self.__class__.__name__)
.add_self_request(self)
.add(
estimator=self.estimator,
method_mapping=MethodMapping().add(caller="fit", callee="fit"),
)
)
return router
class MetaTransformer(MetaEstimatorMixin, TransformerMixin, BaseEstimator):
"""A simple meta-transformer."""
def __init__(self, transformer):
self.transformer = transformer
def fit(self, X, y=None, **fit_params):
params = process_routing(self, "fit", **fit_params)
self.transformer_ = clone(self.transformer).fit(X, y, **params.transformer.fit)
return self
def transform(self, X, y=None, **transform_params):
params = process_routing(self, "transform", **transform_params)
return self.transformer_.transform(X, **params.transformer.transform)
def get_metadata_routing(self):
return MetadataRouter(owner=self.__class__.__name__).add(
transformer=self.transformer,
method_mapping=MethodMapping()
.add(caller="fit", callee="fit")
.add(caller="transform", callee="transform"),
)
|