laywerrobot/lib/python3.6/site-packages/sklearn/utils/estimator_checks.py

1753 lines
65 KiB
Python
Raw Normal View History

2020-08-27 21:55:39 +02:00
from __future__ import print_function
import types
import warnings
import sys
import traceback
import pickle
from copy import deepcopy
import numpy as np
from scipy import sparse
from scipy.stats import rankdata
import struct
from sklearn.externals.six.moves import zip
from sklearn.externals.joblib import hash, Memory
from sklearn.utils.testing import assert_raises
from sklearn.utils.testing import assert_raises_regex
from sklearn.utils.testing import assert_raise_message
from sklearn.utils.testing import assert_equal
from sklearn.utils.testing import assert_not_equal
from sklearn.utils.testing import assert_true
from sklearn.utils.testing import assert_false
from sklearn.utils.testing import assert_in
from sklearn.utils.testing import assert_array_equal
from sklearn.utils.testing import assert_allclose
from sklearn.utils.testing import assert_allclose_dense_sparse
from sklearn.utils.testing import assert_warns_message
from sklearn.utils.testing import META_ESTIMATORS
from sklearn.utils.testing import set_random_state
from sklearn.utils.testing import assert_greater
from sklearn.utils.testing import assert_greater_equal
from sklearn.utils.testing import SkipTest
from sklearn.utils.testing import ignore_warnings
from sklearn.utils.testing import assert_dict_equal
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.base import (clone, TransformerMixin, ClusterMixin,
BaseEstimator, is_classifier, is_regressor)
from sklearn.metrics import accuracy_score, adjusted_rand_score, f1_score
from sklearn.random_projection import BaseRandomProjection
from sklearn.feature_selection import SelectKBest
from sklearn.svm.base import BaseLibSVM
from sklearn.linear_model.stochastic_gradient import BaseSGD
from sklearn.pipeline import make_pipeline
from sklearn.exceptions import ConvergenceWarning
from sklearn.exceptions import DataConversionWarning
from sklearn.exceptions import SkipTestWarning
from sklearn.model_selection import train_test_split
from sklearn.utils import shuffle
from sklearn.utils.fixes import signature
from sklearn.utils.validation import has_fit_parameter, _num_samples
from sklearn.preprocessing import StandardScaler
from sklearn.datasets import load_iris, load_boston, make_blobs
BOSTON = None
CROSS_DECOMPOSITION = ['PLSCanonical', 'PLSRegression', 'CCA', 'PLSSVD']
MULTI_OUTPUT = ['CCA', 'DecisionTreeRegressor', 'ElasticNet',
'ExtraTreeRegressor', 'ExtraTreesRegressor', 'GaussianProcess',
'GaussianProcessRegressor',
'KNeighborsRegressor', 'KernelRidge', 'Lars', 'Lasso',
'LassoLars', 'LinearRegression', 'MultiTaskElasticNet',
'MultiTaskElasticNetCV', 'MultiTaskLasso', 'MultiTaskLassoCV',
'OrthogonalMatchingPursuit', 'PLSCanonical', 'PLSRegression',
'RANSACRegressor', 'RadiusNeighborsRegressor',
'RandomForestRegressor', 'Ridge', 'RidgeCV']
def _yield_non_meta_checks(name, estimator):
yield check_estimators_dtypes
yield check_fit_score_takes_y
yield check_dtype_object
yield check_sample_weights_pandas_series
yield check_sample_weights_list
yield check_estimators_fit_returns_self
# Check that all estimator yield informative messages when
# trained on empty datasets
yield check_estimators_empty_data_messages
if name not in CROSS_DECOMPOSITION + ['SpectralEmbedding']:
# SpectralEmbedding is non-deterministic,
# see issue #4236
# cross-decomposition's "transform" returns X and Y
yield check_pipeline_consistency
if name not in ['Imputer']:
# Test that all estimators check their input for NaN's and infs
yield check_estimators_nan_inf
if name not in ['GaussianProcess']:
# FIXME!
# in particular GaussianProcess!
yield check_estimators_overwrite_params
if hasattr(estimator, 'sparsify'):
yield check_sparsify_coefficients
yield check_estimator_sparse_data
# Test that estimators can be pickled, and once pickled
# give the same answer as before.
yield check_estimators_pickle
def _yield_classifier_checks(name, classifier):
# test classifiers can handle non-array data
yield check_classifier_data_not_an_array
# test classifiers trained on a single label always return this label
yield check_classifiers_one_label
yield check_classifiers_classes
yield check_estimators_partial_fit_n_features
# basic consistency testing
yield check_classifiers_train
yield check_classifiers_regression_target
if (name not in
["MultinomialNB", "LabelPropagation", "LabelSpreading"] and
# TODO some complication with -1 label
name not in ["DecisionTreeClassifier", "ExtraTreeClassifier"]):
# We don't raise a warning in these classifiers, as
# the column y interface is used by the forests.
yield check_supervised_y_2d
# test if NotFittedError is raised
yield check_estimators_unfitted
if 'class_weight' in classifier.get_params().keys():
yield check_class_weight_classifiers
yield check_non_transformer_estimators_n_iter
# test if predict_proba is a monotonic transformation of decision_function
yield check_decision_proba_consistency
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_supervised_y_no_nan(name, estimator_orig):
# Checks that the Estimator targets are not NaN.
estimator = clone(estimator_orig)
rng = np.random.RandomState(888)
X = rng.randn(10, 5)
y = np.ones(10) * np.inf
y = multioutput_estimator_convert_y_2d(estimator, y)
errmsg = "Input contains NaN, infinity or a value too large for " \
"dtype('float64')."
try:
estimator.fit(X, y)
except ValueError as e:
if str(e) != errmsg:
raise ValueError("Estimator {0} raised error as expected, but "
"does not match expected error message"
.format(name))
else:
raise ValueError("Estimator {0} should have raised error on fitting "
"array y with NaN value.".format(name))
def _yield_regressor_checks(name, regressor):
# TODO: test with intercept
# TODO: test with multiple responses
# basic testing
yield check_regressors_train
yield check_regressor_data_not_an_array
yield check_estimators_partial_fit_n_features
yield check_regressors_no_decision_function
yield check_supervised_y_2d
yield check_supervised_y_no_nan
if name != 'CCA':
# check that the regressor handles int input
yield check_regressors_int
if name != "GaussianProcessRegressor":
# Test if NotFittedError is raised
yield check_estimators_unfitted
yield check_non_transformer_estimators_n_iter
def _yield_transformer_checks(name, transformer):
# All transformers should either deal with sparse data or raise an
# exception with type TypeError and an intelligible error message
if name not in ['AdditiveChi2Sampler', 'Binarizer', 'Normalizer',
'PLSCanonical', 'PLSRegression', 'CCA', 'PLSSVD']:
yield check_transformer_data_not_an_array
# these don't actually fit the data, so don't raise errors
if name not in ['AdditiveChi2Sampler', 'Binarizer',
'FunctionTransformer', 'Normalizer']:
# basic tests
yield check_transformer_general
yield check_transformers_unfitted
# Dependent on external solvers and hence accessing the iter
# param is non-trivial.
external_solver = ['Isomap', 'KernelPCA', 'LocallyLinearEmbedding',
'RandomizedLasso', 'LogisticRegressionCV']
if name not in external_solver:
yield check_transformer_n_iter
def _yield_clustering_checks(name, clusterer):
yield check_clusterer_compute_labels_predict
if name not in ('WardAgglomeration', "FeatureAgglomeration"):
# this is clustering on the features
# let's not test that here.
yield check_clustering
yield check_estimators_partial_fit_n_features
yield check_non_transformer_estimators_n_iter
def _yield_all_checks(name, estimator):
for check in _yield_non_meta_checks(name, estimator):
yield check
if is_classifier(estimator):
for check in _yield_classifier_checks(name, estimator):
yield check
if is_regressor(estimator):
for check in _yield_regressor_checks(name, estimator):
yield check
if isinstance(estimator, TransformerMixin):
for check in _yield_transformer_checks(name, estimator):
yield check
if isinstance(estimator, ClusterMixin):
for check in _yield_clustering_checks(name, estimator):
yield check
yield check_fit2d_predict1d
yield check_fit2d_1sample
yield check_fit2d_1feature
yield check_fit1d_1feature
yield check_fit1d_1sample
yield check_get_params_invariance
yield check_dict_unchanged
yield check_dont_overwrite_parameters
def check_estimator(Estimator):
"""Check if estimator adheres to scikit-learn conventions.
This estimator will run an extensive test-suite for input validation,
shapes, etc.
Additional tests for classifiers, regressors, clustering or transformers
will be run if the Estimator class inherits from the corresponding mixin
from sklearn.base.
This test can be applied to classes or instances.
Classes currently have some additional tests that related to construction,
while passing instances allows the testing of multiple options.
Parameters
----------
estimator : estimator object or class
Estimator to check. Estimator is a class object or instance.
"""
if isinstance(Estimator, type):
# got a class
name = Estimator.__name__
check_parameters_default_constructible(name, Estimator)
check_no_fit_attributes_set_in_init(name, Estimator)
estimator = Estimator()
else:
# got an instance
estimator = Estimator
name = type(estimator).__name__
for check in _yield_all_checks(name, estimator):
try:
check(name, estimator)
except SkipTest as message:
# the only SkipTest thrown currently results from not
# being able to import pandas.
warnings.warn(message, SkipTestWarning)
def _boston_subset(n_samples=200):
global BOSTON
if BOSTON is None:
boston = load_boston()
X, y = boston.data, boston.target
X, y = shuffle(X, y, random_state=0)
X, y = X[:n_samples], y[:n_samples]
X = StandardScaler().fit_transform(X)
BOSTON = X, y
return BOSTON
def set_checking_parameters(estimator):
# set parameters to speed up some estimators and
# avoid deprecated behaviour
params = estimator.get_params()
if ("n_iter" in params and estimator.__class__.__name__ != "TSNE"
and not isinstance(estimator, BaseSGD)):
estimator.set_params(n_iter=5)
if "max_iter" in params:
warnings.simplefilter("ignore", ConvergenceWarning)
if estimator.max_iter is not None:
estimator.set_params(max_iter=min(5, estimator.max_iter))
# LinearSVR, LinearSVC
if estimator.__class__.__name__ in ['LinearSVR', 'LinearSVC']:
estimator.set_params(max_iter=20)
# NMF
if estimator.__class__.__name__ == 'NMF':
estimator.set_params(max_iter=100)
# MLP
if estimator.__class__.__name__ in ['MLPClassifier', 'MLPRegressor']:
estimator.set_params(max_iter=100)
if "n_resampling" in params:
# randomized lasso
estimator.set_params(n_resampling=5)
if "n_estimators" in params:
# especially gradient boosting with default 100
estimator.set_params(n_estimators=min(5, estimator.n_estimators))
if "max_trials" in params:
# RANSAC
estimator.set_params(max_trials=10)
if "n_init" in params:
# K-Means
estimator.set_params(n_init=2)
if "decision_function_shape" in params:
# SVC
estimator.set_params(decision_function_shape='ovo')
if estimator.__class__.__name__ == "SelectFdr":
# be tolerant of noisy datasets (not actually speed)
estimator.set_params(alpha=.5)
if estimator.__class__.__name__ == "TheilSenRegressor":
estimator.max_subpopulation = 100
if isinstance(estimator, BaseRandomProjection):
# Due to the jl lemma and often very few samples, the number
# of components of the random matrix projection will be probably
# greater than the number of features.
# So we impose a smaller number (avoid "auto" mode)
estimator.set_params(n_components=2)
if isinstance(estimator, SelectKBest):
# SelectKBest has a default of k=10
# which is more feature than we have in most case.
estimator.set_params(k=1)
class NotAnArray(object):
" An object that is convertable to an array"
def __init__(self, data):
self.data = data
def __array__(self, dtype=None):
return self.data
def _is_32bit():
"""Detect if process is 32bit Python."""
return struct.calcsize('P') * 8 == 32
def check_estimator_sparse_data(name, estimator_orig):
rng = np.random.RandomState(0)
X = rng.rand(40, 10)
X[X < .8] = 0
X_csr = sparse.csr_matrix(X)
y = (4 * rng.rand(40)).astype(np.int)
# catch deprecation warnings
with ignore_warnings(category=DeprecationWarning):
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
for sparse_format in ['csr', 'csc', 'dok', 'lil', 'coo', 'dia', 'bsr']:
X = X_csr.asformat(sparse_format)
# catch deprecation warnings
with ignore_warnings(category=(DeprecationWarning, FutureWarning)):
if name in ['Scaler', 'StandardScaler']:
estimator = clone(estimator).set_params(with_mean=False)
else:
estimator = clone(estimator)
# fit and predict
try:
with ignore_warnings(category=(DeprecationWarning, FutureWarning)):
estimator.fit(X, y)
if hasattr(estimator, "predict"):
pred = estimator.predict(X)
assert_equal(pred.shape, (X.shape[0],))
if hasattr(estimator, 'predict_proba'):
probs = estimator.predict_proba(X)
assert_equal(probs.shape, (X.shape[0], 4))
except TypeError as e:
if 'sparse' not in repr(e):
print("Estimator %s doesn't seem to fail gracefully on "
"sparse data: error message state explicitly that "
"sparse input is not supported if this is not the case."
% name)
raise
except Exception:
print("Estimator %s doesn't seem to fail gracefully on "
"sparse data: it should raise a TypeError if sparse input "
"is explicitly not supported." % name)
raise
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_sample_weights_pandas_series(name, estimator_orig):
# check that estimators will accept a 'sample_weight' parameter of
# type pandas.Series in the 'fit' function.
estimator = clone(estimator_orig)
if has_fit_parameter(estimator, "sample_weight"):
try:
import pandas as pd
X = pd.DataFrame([[1, 1], [1, 2], [1, 3], [2, 1], [2, 2], [2, 3]])
y = pd.Series([1, 1, 1, 2, 2, 2])
weights = pd.Series([1] * 6)
try:
estimator.fit(X, y, sample_weight=weights)
except ValueError:
raise ValueError("Estimator {0} raises error if "
"'sample_weight' parameter is of "
"type pandas.Series".format(name))
except ImportError:
raise SkipTest("pandas is not installed: not testing for "
"input of type pandas.Series to class weight.")
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_sample_weights_list(name, estimator_orig):
# check that estimators will accept a 'sample_weight' parameter of
# type list in the 'fit' function.
if has_fit_parameter(estimator_orig, "sample_weight"):
estimator = clone(estimator_orig)
rnd = np.random.RandomState(0)
X = rnd.uniform(size=(10, 3))
y = np.arange(10) % 3
y = multioutput_estimator_convert_y_2d(estimator, y)
sample_weight = [3] * 10
# Test that estimators don't raise any exception
estimator.fit(X, y, sample_weight=sample_weight)
@ignore_warnings(category=(DeprecationWarning, FutureWarning, UserWarning))
def check_dtype_object(name, estimator_orig):
# check that estimators treat dtype object as numeric if possible
rng = np.random.RandomState(0)
X = rng.rand(40, 10).astype(object)
y = (X[:, 0] * 4).astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
estimator.fit(X, y)
if hasattr(estimator, "predict"):
estimator.predict(X)
if hasattr(estimator, "transform"):
estimator.transform(X)
try:
estimator.fit(X, y.astype(object))
except Exception as e:
if "Unknown label type" not in str(e):
raise
X[0, 0] = {'foo': 'bar'}
msg = "argument must be a string or a number"
assert_raises_regex(TypeError, msg, estimator.fit, X, y)
@ignore_warnings
def check_dict_unchanged(name, estimator_orig):
# this estimator raises
# ValueError: Found array with 0 feature(s) (shape=(23, 0))
# while a minimum of 1 is required.
# error
if name in ['SpectralCoclustering']:
return
rnd = np.random.RandomState(0)
if name in ['RANSACRegressor']:
X = 3 * rnd.uniform(size=(20, 3))
else:
X = 2 * rnd.uniform(size=(20, 3))
y = X[:, 0].astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
if hasattr(estimator, "n_best"):
estimator.n_best = 1
set_random_state(estimator, 1)
estimator.fit(X, y)
for method in ["predict", "transform", "decision_function",
"predict_proba"]:
if hasattr(estimator, method):
dict_before = estimator.__dict__.copy()
getattr(estimator, method)(X)
assert_dict_equal(estimator.__dict__, dict_before,
'Estimator changes __dict__ during %s' % method)
def is_public_parameter(attr):
return not (attr.startswith('_') or attr.endswith('_'))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_dont_overwrite_parameters(name, estimator_orig):
# check that fit method only changes or sets private attributes
if hasattr(estimator_orig.__init__, "deprecated_original"):
# to not check deprecated classes
return
estimator = clone(estimator_orig)
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(20, 3))
y = X[:, 0].astype(np.int)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
dict_before_fit = estimator.__dict__.copy()
estimator.fit(X, y)
dict_after_fit = estimator.__dict__
public_keys_after_fit = [key for key in dict_after_fit.keys()
if is_public_parameter(key)]
attrs_added_by_fit = [key for key in public_keys_after_fit
if key not in dict_before_fit.keys()]
# check that fit doesn't add any public attribute
assert_true(not attrs_added_by_fit,
('Estimator adds public attribute(s) during'
' the fit method.'
' Estimators are only allowed to add private attributes'
' either started with _ or ended'
' with _ but %s added' % ', '.join(attrs_added_by_fit)))
# check that fit doesn't change any public attribute
attrs_changed_by_fit = [key for key in public_keys_after_fit
if (dict_before_fit[key]
is not dict_after_fit[key])]
assert_true(not attrs_changed_by_fit,
('Estimator changes public attribute(s) during'
' the fit method. Estimators are only allowed'
' to change attributes started'
' or ended with _, but'
' %s changed' % ', '.join(attrs_changed_by_fit)))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_fit2d_predict1d(name, estimator_orig):
# check by fitting a 2d array and predicting with a 1d array
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(20, 3))
y = X[:, 0].astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
estimator.fit(X, y)
for method in ["predict", "transform", "decision_function",
"predict_proba"]:
if hasattr(estimator, method):
assert_raise_message(ValueError, "Reshape your data",
getattr(estimator, method), X[0])
@ignore_warnings
def check_fit2d_1sample(name, estimator_orig):
# check by fitting a 2d array and prediting with a 1d array
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(1, 10))
y = X[:, 0].astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
try:
estimator.fit(X, y)
except ValueError:
pass
@ignore_warnings
def check_fit2d_1feature(name, estimator_orig):
# check by fitting a 2d array and prediting with a 1d array
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(10, 1))
y = X[:, 0].astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
try:
estimator.fit(X, y)
except ValueError:
pass
@ignore_warnings
def check_fit1d_1feature(name, estimator_orig):
# check fitting 1d array with 1 feature
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(20))
y = X.astype(np.int)
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
try:
estimator.fit(X, y)
except ValueError:
pass
@ignore_warnings
def check_fit1d_1sample(name, estimator_orig):
# check fitting 1d array with 1 feature
rnd = np.random.RandomState(0)
X = 3 * rnd.uniform(size=(20))
y = np.array([1])
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
if hasattr(estimator, "n_components"):
estimator.n_components = 1
if hasattr(estimator, "n_clusters"):
estimator.n_clusters = 1
set_random_state(estimator, 1)
try:
estimator.fit(X, y)
except ValueError:
pass
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_transformer_general(name, transformer):
X, y = make_blobs(n_samples=30, centers=[[0, 0, 0], [1, 1, 1]],
random_state=0, n_features=2, cluster_std=0.1)
X = StandardScaler().fit_transform(X)
X -= X.min()
_check_transformer(name, transformer, X, y)
_check_transformer(name, transformer, X.tolist(), y.tolist())
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_transformer_data_not_an_array(name, transformer):
X, y = make_blobs(n_samples=30, centers=[[0, 0, 0], [1, 1, 1]],
random_state=0, n_features=2, cluster_std=0.1)
X = StandardScaler().fit_transform(X)
# We need to make sure that we have non negative data, for things
# like NMF
X -= X.min() - .1
this_X = NotAnArray(X)
this_y = NotAnArray(np.asarray(y))
_check_transformer(name, transformer, this_X, this_y)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_transformers_unfitted(name, transformer):
X, y = _boston_subset()
transformer = clone(transformer)
assert_raises((AttributeError, ValueError), transformer.transform, X)
def _check_transformer(name, transformer_orig, X, y):
if name in ('CCA', 'LocallyLinearEmbedding', 'KernelPCA') and _is_32bit():
# Those transformers yield non-deterministic output when executed on
# a 32bit Python. The same transformers are stable on 64bit Python.
# FIXME: try to isolate a minimalistic reproduction case only depending
# on numpy & scipy and/or maybe generate a test dataset that does not
# cause such unstable behaviors.
msg = name + ' is non deterministic on 32bit Python'
raise SkipTest(msg)
n_samples, n_features = np.asarray(X).shape
transformer = clone(transformer_orig)
set_random_state(transformer)
# fit
if name in CROSS_DECOMPOSITION:
y_ = np.c_[y, y]
y_[::2, 1] *= 2
else:
y_ = y
transformer.fit(X, y_)
# fit_transform method should work on non fitted estimator
transformer_clone = clone(transformer)
X_pred = transformer_clone.fit_transform(X, y=y_)
if isinstance(X_pred, tuple):
for x_pred in X_pred:
assert_equal(x_pred.shape[0], n_samples)
else:
# check for consistent n_samples
assert_equal(X_pred.shape[0], n_samples)
if hasattr(transformer, 'transform'):
if name in CROSS_DECOMPOSITION:
X_pred2 = transformer.transform(X, y_)
X_pred3 = transformer.fit_transform(X, y=y_)
else:
X_pred2 = transformer.transform(X)
X_pred3 = transformer.fit_transform(X, y=y_)
if isinstance(X_pred, tuple) and isinstance(X_pred2, tuple):
for x_pred, x_pred2, x_pred3 in zip(X_pred, X_pred2, X_pred3):
assert_allclose_dense_sparse(
x_pred, x_pred2, atol=1e-2,
err_msg="fit_transform and transform outcomes "
"not consistent in %s"
% transformer)
assert_allclose_dense_sparse(
x_pred, x_pred3, atol=1e-2,
err_msg="consecutive fit_transform outcomes "
"not consistent in %s"
% transformer)
else:
assert_allclose_dense_sparse(
X_pred, X_pred2,
err_msg="fit_transform and transform outcomes "
"not consistent in %s"
% transformer, atol=1e-2)
assert_allclose_dense_sparse(
X_pred, X_pred3, atol=1e-2,
err_msg="consecutive fit_transform outcomes "
"not consistent in %s"
% transformer)
assert_equal(_num_samples(X_pred2), n_samples)
assert_equal(_num_samples(X_pred3), n_samples)
# raises error on malformed input for transform
if hasattr(X, 'T'):
# If it's not an array, it does not have a 'T' property
assert_raises(ValueError, transformer.transform, X.T)
@ignore_warnings
def check_pipeline_consistency(name, estimator_orig):
if name in ('CCA', 'LocallyLinearEmbedding', 'KernelPCA') and _is_32bit():
# Those transformers yield non-deterministic output when executed on
# a 32bit Python. The same transformers are stable on 64bit Python.
# FIXME: try to isolate a minimalistic reproduction case only depending
# scipy and/or maybe generate a test dataset that does not
# cause such unstable behaviors.
msg = name + ' is non deterministic on 32bit Python'
raise SkipTest(msg)
# check that make_pipeline(est) gives same score as est
X, y = make_blobs(n_samples=30, centers=[[0, 0, 0], [1, 1, 1]],
random_state=0, n_features=2, cluster_std=0.1)
X -= X.min()
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
set_random_state(estimator)
pipeline = make_pipeline(estimator)
estimator.fit(X, y)
pipeline.fit(X, y)
funcs = ["score", "fit_transform"]
for func_name in funcs:
func = getattr(estimator, func_name, None)
if func is not None:
func_pipeline = getattr(pipeline, func_name)
result = func(X, y)
result_pipe = func_pipeline(X, y)
assert_allclose_dense_sparse(result, result_pipe)
@ignore_warnings
def check_fit_score_takes_y(name, estimator_orig):
# check that all estimators accept an optional y
# in fit and score so they can be used in pipelines
rnd = np.random.RandomState(0)
X = rnd.uniform(size=(10, 3))
y = np.arange(10) % 3
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
set_random_state(estimator)
funcs = ["fit", "score", "partial_fit", "fit_predict", "fit_transform"]
for func_name in funcs:
func = getattr(estimator, func_name, None)
if func is not None:
func(X, y)
args = [p.name for p in signature(func).parameters.values()]
if args[0] == "self":
# if_delegate_has_method makes methods into functions
# with an explicit "self", so need to shift arguments
args = args[1:]
assert_true(args[1] in ["y", "Y"],
"Expected y or Y as second argument for method "
"%s of %s. Got arguments: %r."
% (func_name, type(estimator).__name__, args))
@ignore_warnings
def check_estimators_dtypes(name, estimator_orig):
rnd = np.random.RandomState(0)
X_train_32 = 3 * rnd.uniform(size=(20, 5)).astype(np.float32)
X_train_64 = X_train_32.astype(np.float64)
X_train_int_64 = X_train_32.astype(np.int64)
X_train_int_32 = X_train_32.astype(np.int32)
y = X_train_int_64[:, 0]
y = multioutput_estimator_convert_y_2d(estimator_orig, y)
methods = ["predict", "transform", "decision_function", "predict_proba"]
for X_train in [X_train_32, X_train_64, X_train_int_64, X_train_int_32]:
estimator = clone(estimator_orig)
set_random_state(estimator, 1)
estimator.fit(X_train, y)
for method in methods:
if hasattr(estimator, method):
getattr(estimator, method)(X_train)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_estimators_empty_data_messages(name, estimator_orig):
e = clone(estimator_orig)
set_random_state(e, 1)
X_zero_samples = np.empty(0).reshape(0, 3)
# The precise message can change depending on whether X or y is
# validated first. Let us test the type of exception only:
assert_raises(ValueError, e.fit, X_zero_samples, [])
X_zero_features = np.empty(0).reshape(3, 0)
# the following y should be accepted by both classifiers and regressors
# and ignored by unsupervised models
y = multioutput_estimator_convert_y_2d(e, np.array([1, 0, 1]))
msg = ("0 feature\(s\) \(shape=\(3, 0\)\) while a minimum of \d* "
"is required.")
assert_raises_regex(ValueError, msg, e.fit, X_zero_features, y)
@ignore_warnings(category=DeprecationWarning)
def check_estimators_nan_inf(name, estimator_orig):
# Checks that Estimator X's do not contain NaN or inf.
rnd = np.random.RandomState(0)
X_train_finite = rnd.uniform(size=(10, 3))
X_train_nan = rnd.uniform(size=(10, 3))
X_train_nan[0, 0] = np.nan
X_train_inf = rnd.uniform(size=(10, 3))
X_train_inf[0, 0] = np.inf
y = np.ones(10)
y[:5] = 0
y = multioutput_estimator_convert_y_2d(estimator_orig, y)
error_string_fit = "Estimator doesn't check for NaN and inf in fit."
error_string_predict = ("Estimator doesn't check for NaN and inf in"
" predict.")
error_string_transform = ("Estimator doesn't check for NaN and inf in"
" transform.")
for X_train in [X_train_nan, X_train_inf]:
# catch deprecation warnings
with ignore_warnings(category=(DeprecationWarning, FutureWarning)):
estimator = clone(estimator_orig)
set_random_state(estimator, 1)
# try to fit
try:
estimator.fit(X_train, y)
except ValueError as e:
if 'inf' not in repr(e) and 'NaN' not in repr(e):
print(error_string_fit, estimator, e)
traceback.print_exc(file=sys.stdout)
raise e
except Exception as exc:
print(error_string_fit, estimator, exc)
traceback.print_exc(file=sys.stdout)
raise exc
else:
raise AssertionError(error_string_fit, estimator)
# actually fit
estimator.fit(X_train_finite, y)
# predict
if hasattr(estimator, "predict"):
try:
estimator.predict(X_train)
except ValueError as e:
if 'inf' not in repr(e) and 'NaN' not in repr(e):
print(error_string_predict, estimator, e)
traceback.print_exc(file=sys.stdout)
raise e
except Exception as exc:
print(error_string_predict, estimator, exc)
traceback.print_exc(file=sys.stdout)
else:
raise AssertionError(error_string_predict, estimator)
# transform
if hasattr(estimator, "transform"):
try:
estimator.transform(X_train)
except ValueError as e:
if 'inf' not in repr(e) and 'NaN' not in repr(e):
print(error_string_transform, estimator, e)
traceback.print_exc(file=sys.stdout)
raise e
except Exception as exc:
print(error_string_transform, estimator, exc)
traceback.print_exc(file=sys.stdout)
else:
raise AssertionError(error_string_transform, estimator)
@ignore_warnings
def check_estimators_pickle(name, estimator_orig):
"""Test that we can pickle all estimators"""
check_methods = ["predict", "transform", "decision_function",
"predict_proba"]
X, y = make_blobs(n_samples=30, centers=[[0, 0, 0], [1, 1, 1]],
random_state=0, n_features=2, cluster_std=0.1)
# some estimators can't do features less than 0
X -= X.min()
estimator = clone(estimator_orig)
# some estimators only take multioutputs
y = multioutput_estimator_convert_y_2d(estimator, y)
set_random_state(estimator)
estimator.fit(X, y)
result = dict()
for method in check_methods:
if hasattr(estimator, method):
result[method] = getattr(estimator, method)(X)
# pickle and unpickle!
pickled_estimator = pickle.dumps(estimator)
if estimator.__module__.startswith('sklearn.'):
assert_true(b"version" in pickled_estimator)
unpickled_estimator = pickle.loads(pickled_estimator)
for method in result:
unpickled_result = getattr(unpickled_estimator, method)(X)
assert_allclose_dense_sparse(result[method], unpickled_result)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_estimators_partial_fit_n_features(name, estimator_orig):
# check if number of features changes between calls to partial_fit.
if not hasattr(estimator_orig, 'partial_fit'):
return
estimator = clone(estimator_orig)
X, y = make_blobs(n_samples=50, random_state=1)
X -= X.min()
try:
if is_classifier(estimator):
classes = np.unique(y)
estimator.partial_fit(X, y, classes=classes)
else:
estimator.partial_fit(X, y)
except NotImplementedError:
return
assert_raises(ValueError, estimator.partial_fit, X[:, :-1], y)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_clustering(name, clusterer_orig):
clusterer = clone(clusterer_orig)
X, y = make_blobs(n_samples=50, random_state=1)
X, y = shuffle(X, y, random_state=7)
X = StandardScaler().fit_transform(X)
n_samples, n_features = X.shape
# catch deprecation and neighbors warnings
if hasattr(clusterer, "n_clusters"):
clusterer.set_params(n_clusters=3)
set_random_state(clusterer)
if name == 'AffinityPropagation':
clusterer.set_params(preference=-100)
clusterer.set_params(max_iter=100)
# fit
clusterer.fit(X)
# with lists
clusterer.fit(X.tolist())
assert_equal(clusterer.labels_.shape, (n_samples,))
pred = clusterer.labels_
assert_greater(adjusted_rand_score(pred, y), 0.4)
# fit another time with ``fit_predict`` and compare results
if name == 'SpectralClustering':
# there is no way to make Spectral clustering deterministic :(
return
set_random_state(clusterer)
with warnings.catch_warnings(record=True):
pred2 = clusterer.fit_predict(X)
assert_array_equal(pred, pred2)
@ignore_warnings(category=DeprecationWarning)
def check_clusterer_compute_labels_predict(name, clusterer_orig):
"""Check that predict is invariant of compute_labels"""
X, y = make_blobs(n_samples=20, random_state=0)
clusterer = clone(clusterer_orig)
if hasattr(clusterer, "compute_labels"):
# MiniBatchKMeans
if hasattr(clusterer, "random_state"):
clusterer.set_params(random_state=0)
X_pred1 = clusterer.fit(X).predict(X)
clusterer.set_params(compute_labels=False)
X_pred2 = clusterer.fit(X).predict(X)
assert_array_equal(X_pred1, X_pred2)
@ignore_warnings(category=DeprecationWarning)
def check_classifiers_one_label(name, classifier_orig):
error_string_fit = "Classifier can't train when only one class is present."
error_string_predict = ("Classifier can't predict when only one class is "
"present.")
rnd = np.random.RandomState(0)
X_train = rnd.uniform(size=(10, 3))
X_test = rnd.uniform(size=(10, 3))
y = np.ones(10)
# catch deprecation warnings
with ignore_warnings(category=(DeprecationWarning, FutureWarning)):
classifier = clone(classifier_orig)
# try to fit
try:
classifier.fit(X_train, y)
except ValueError as e:
if 'class' not in repr(e):
print(error_string_fit, classifier, e)
traceback.print_exc(file=sys.stdout)
raise e
else:
return
except Exception as exc:
print(error_string_fit, classifier, exc)
traceback.print_exc(file=sys.stdout)
raise exc
# predict
try:
assert_array_equal(classifier.predict(X_test), y)
except Exception as exc:
print(error_string_predict, classifier, exc)
raise exc
@ignore_warnings # Warnings are raised by decision function
def check_classifiers_train(name, classifier_orig):
X_m, y_m = make_blobs(n_samples=300, random_state=0)
X_m, y_m = shuffle(X_m, y_m, random_state=7)
X_m = StandardScaler().fit_transform(X_m)
# generate binary problem from multi-class one
y_b = y_m[y_m != 2]
X_b = X_m[y_m != 2]
for (X, y) in [(X_m, y_m), (X_b, y_b)]:
classes = np.unique(y)
n_classes = len(classes)
n_samples, n_features = X.shape
classifier = clone(classifier_orig)
if name in ['BernoulliNB', 'MultinomialNB']:
X -= X.min()
set_random_state(classifier)
# raises error on malformed input for fit
assert_raises(ValueError, classifier.fit, X, y[:-1])
# fit
classifier.fit(X, y)
# with lists
classifier.fit(X.tolist(), y.tolist())
assert_true(hasattr(classifier, "classes_"))
y_pred = classifier.predict(X)
assert_equal(y_pred.shape, (n_samples,))
# training set performance
if name not in ['BernoulliNB', 'MultinomialNB']:
assert_greater(accuracy_score(y, y_pred), 0.83)
# raises error on malformed input for predict
assert_raises(ValueError, classifier.predict, X.T)
if hasattr(classifier, "decision_function"):
try:
# decision_function agrees with predict
decision = classifier.decision_function(X)
if n_classes == 2:
assert_equal(decision.shape, (n_samples,))
dec_pred = (decision.ravel() > 0).astype(np.int)
assert_array_equal(dec_pred, y_pred)
if (n_classes == 3 and
# 1on1 of LibSVM works differently
not isinstance(classifier, BaseLibSVM)):
assert_equal(decision.shape, (n_samples, n_classes))
assert_array_equal(np.argmax(decision, axis=1), y_pred)
# raises error on malformed input
assert_raises(ValueError,
classifier.decision_function, X.T)
# raises error on malformed input for decision_function
assert_raises(ValueError,
classifier.decision_function, X.T)
except NotImplementedError:
pass
if hasattr(classifier, "predict_proba"):
# predict_proba agrees with predict
y_prob = classifier.predict_proba(X)
assert_equal(y_prob.shape, (n_samples, n_classes))
assert_array_equal(np.argmax(y_prob, axis=1), y_pred)
# check that probas for all classes sum to one
assert_allclose(np.sum(y_prob, axis=1), np.ones(n_samples))
# raises error on malformed input
assert_raises(ValueError, classifier.predict_proba, X.T)
# raises error on malformed input for predict_proba
assert_raises(ValueError, classifier.predict_proba, X.T)
if hasattr(classifier, "predict_log_proba"):
# predict_log_proba is a transformation of predict_proba
y_log_prob = classifier.predict_log_proba(X)
assert_allclose(y_log_prob, np.log(y_prob), 8, atol=1e-9)
assert_array_equal(np.argsort(y_log_prob), np.argsort(y_prob))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_estimators_fit_returns_self(name, estimator_orig):
"""Check if self is returned when calling fit"""
X, y = make_blobs(random_state=0, n_samples=9, n_features=4)
# some want non-negative input
X -= X.min()
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
set_random_state(estimator)
assert_true(estimator.fit(X, y) is estimator)
@ignore_warnings
def check_estimators_unfitted(name, estimator_orig):
"""Check that predict raises an exception in an unfitted estimator.
Unfitted estimators should raise either AttributeError or ValueError.
The specific exception type NotFittedError inherits from both and can
therefore be adequately raised for that purpose.
"""
# Common test for Regressors as well as Classifiers
X, y = _boston_subset()
est = clone(estimator_orig)
msg = "fit"
if hasattr(est, 'predict'):
assert_raise_message((AttributeError, ValueError), msg,
est.predict, X)
if hasattr(est, 'decision_function'):
assert_raise_message((AttributeError, ValueError), msg,
est.decision_function, X)
if hasattr(est, 'predict_proba'):
assert_raise_message((AttributeError, ValueError), msg,
est.predict_proba, X)
if hasattr(est, 'predict_log_proba'):
assert_raise_message((AttributeError, ValueError), msg,
est.predict_log_proba, X)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_supervised_y_2d(name, estimator_orig):
if "MultiTask" in name:
# These only work on 2d, so this test makes no sense
return
if name == "GaussianProcess":
# Workaround: https://github.com/scikit-learn/scikit-learn/issues/10562
return
rnd = np.random.RandomState(0)
X = rnd.uniform(size=(10, 3))
y = np.arange(10) % 3
estimator = clone(estimator_orig)
set_random_state(estimator)
# fit
estimator.fit(X, y)
y_pred = estimator.predict(X)
set_random_state(estimator)
# Check that when a 2D y is given, a DataConversionWarning is
# raised
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always", DataConversionWarning)
warnings.simplefilter("ignore", RuntimeWarning)
estimator.fit(X, y[:, np.newaxis])
y_pred_2d = estimator.predict(X)
msg = "expected 1 DataConversionWarning, got: %s" % (
", ".join([str(w_x) for w_x in w]))
if name not in MULTI_OUTPUT:
# check that we warned if we don't support multi-output
assert_greater(len(w), 0, msg)
assert_true("DataConversionWarning('A column-vector y"
" was passed when a 1d array was expected" in msg)
assert_allclose(y_pred.ravel(), y_pred_2d.ravel())
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_classifiers_classes(name, classifier_orig):
X, y = make_blobs(n_samples=30, random_state=0, cluster_std=0.1)
X, y = shuffle(X, y, random_state=7)
X = StandardScaler().fit_transform(X)
# We need to make sure that we have non negative data, for things
# like NMF
X -= X.min() - .1
y_names = np.array(["one", "two", "three"])[y]
for y_names in [y_names, y_names.astype('O')]:
if name in ["LabelPropagation", "LabelSpreading"]:
# TODO some complication with -1 label
y_ = y
else:
y_ = y_names
classes = np.unique(y_)
classifier = clone(classifier_orig)
if name == 'BernoulliNB':
classifier.set_params(binarize=X.mean())
set_random_state(classifier)
# fit
classifier.fit(X, y_)
y_pred = classifier.predict(X)
# training set performance
assert_array_equal(np.unique(y_), np.unique(y_pred))
if np.any(classifier.classes_ != classes):
print("Unexpected classes_ attribute for %r: "
"expected %s, got %s" %
(classifier, classes, classifier.classes_))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_regressors_int(name, regressor_orig):
X, _ = _boston_subset()
X = X[:50]
rnd = np.random.RandomState(0)
y = rnd.randint(3, size=X.shape[0])
y = multioutput_estimator_convert_y_2d(regressor_orig, y)
rnd = np.random.RandomState(0)
# separate estimators to control random seeds
regressor_1 = clone(regressor_orig)
regressor_2 = clone(regressor_orig)
set_random_state(regressor_1)
set_random_state(regressor_2)
if name in CROSS_DECOMPOSITION:
y_ = np.vstack([y, 2 * y + rnd.randint(2, size=len(y))])
y_ = y_.T
else:
y_ = y
# fit
regressor_1.fit(X, y_)
pred1 = regressor_1.predict(X)
regressor_2.fit(X, y_.astype(np.float))
pred2 = regressor_2.predict(X)
assert_allclose(pred1, pred2, atol=1e-2, err_msg=name)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_regressors_train(name, regressor_orig):
X, y = _boston_subset()
y = StandardScaler().fit_transform(y.reshape(-1, 1)) # X is already scaled
y = y.ravel()
regressor = clone(regressor_orig)
y = multioutput_estimator_convert_y_2d(regressor, y)
rnd = np.random.RandomState(0)
if not hasattr(regressor, 'alphas') and hasattr(regressor, 'alpha'):
# linear regressors need to set alpha, but not generalized CV ones
regressor.alpha = 0.01
if name == 'PassiveAggressiveRegressor':
regressor.C = 0.01
# raises error on malformed input for fit
assert_raises(ValueError, regressor.fit, X, y[:-1])
# fit
if name in CROSS_DECOMPOSITION:
y_ = np.vstack([y, 2 * y + rnd.randint(2, size=len(y))])
y_ = y_.T
else:
y_ = y
set_random_state(regressor)
regressor.fit(X, y_)
regressor.fit(X.tolist(), y_.tolist())
y_pred = regressor.predict(X)
assert_equal(y_pred.shape, y_.shape)
# TODO: find out why PLS and CCA fail. RANSAC is random
# and furthermore assumes the presence of outliers, hence
# skipped
if name not in ('PLSCanonical', 'CCA', 'RANSACRegressor'):
assert_greater(regressor.score(X, y_), 0.5)
@ignore_warnings
def check_regressors_no_decision_function(name, regressor_orig):
# checks whether regressors have decision_function or predict_proba
rng = np.random.RandomState(0)
X = rng.normal(size=(10, 4))
regressor = clone(regressor_orig)
y = multioutput_estimator_convert_y_2d(regressor, X[:, 0])
if hasattr(regressor, "n_components"):
# FIXME CCA, PLS is not robust to rank 1 effects
regressor.n_components = 1
regressor.fit(X, y)
funcs = ["decision_function", "predict_proba", "predict_log_proba"]
for func_name in funcs:
func = getattr(regressor, func_name, None)
if func is None:
# doesn't have function
continue
# has function. Should raise deprecation warning
msg = func_name
assert_warns_message(DeprecationWarning, msg, func, X)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_class_weight_classifiers(name, classifier_orig):
if name == "NuSVC":
# the sparse version has a parameter that doesn't do anything
raise SkipTest
if name.endswith("NB"):
# NaiveBayes classifiers have a somewhat different interface.
# FIXME SOON!
raise SkipTest
for n_centers in [2, 3]:
# create a very noisy dataset
X, y = make_blobs(centers=n_centers, random_state=0, cluster_std=20)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.5,
random_state=0)
n_centers = len(np.unique(y_train))
if n_centers == 2:
class_weight = {0: 1000, 1: 0.0001}
else:
class_weight = {0: 1000, 1: 0.0001, 2: 0.0001}
classifier = clone(classifier_orig).set_params(
class_weight=class_weight)
if hasattr(classifier, "n_iter"):
classifier.set_params(n_iter=100)
if hasattr(classifier, "max_iter"):
classifier.set_params(max_iter=1000)
if hasattr(classifier, "min_weight_fraction_leaf"):
classifier.set_params(min_weight_fraction_leaf=0.01)
set_random_state(classifier)
classifier.fit(X_train, y_train)
y_pred = classifier.predict(X_test)
# XXX: Generally can use 0.89 here. On Windows, LinearSVC gets
# 0.88 (Issue #9111)
assert_greater(np.mean(y_pred == 0), 0.87)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_class_weight_balanced_classifiers(name, classifier_orig, X_train,
y_train, X_test, y_test, weights):
classifier = clone(classifier_orig)
if hasattr(classifier, "n_iter"):
classifier.set_params(n_iter=100)
if hasattr(classifier, "max_iter"):
classifier.set_params(max_iter=1000)
set_random_state(classifier)
classifier.fit(X_train, y_train)
y_pred = classifier.predict(X_test)
classifier.set_params(class_weight='balanced')
classifier.fit(X_train, y_train)
y_pred_balanced = classifier.predict(X_test)
assert_greater(f1_score(y_test, y_pred_balanced, average='weighted'),
f1_score(y_test, y_pred, average='weighted'))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_class_weight_balanced_linear_classifier(name, Classifier):
"""Test class weights with non-contiguous class labels."""
# this is run on classes, not instances, though this should be changed
X = np.array([[-1.0, -1.0], [-1.0, 0], [-.8, -1.0],
[1.0, 1.0], [1.0, 0.0]])
y = np.array([1, 1, 1, -1, -1])
classifier = Classifier()
if hasattr(classifier, "n_iter"):
# This is a very small dataset, default n_iter are likely to prevent
# convergence
classifier.set_params(n_iter=1000)
if hasattr(classifier, "max_iter"):
classifier.set_params(max_iter=1000)
set_random_state(classifier)
# Let the model compute the class frequencies
classifier.set_params(class_weight='balanced')
coef_balanced = classifier.fit(X, y).coef_.copy()
# Count each label occurrence to reweight manually
n_samples = len(y)
n_classes = float(len(np.unique(y)))
class_weight = {1: n_samples / (np.sum(y == 1) * n_classes),
-1: n_samples / (np.sum(y == -1) * n_classes)}
classifier.set_params(class_weight=class_weight)
coef_manual = classifier.fit(X, y).coef_.copy()
assert_allclose(coef_balanced, coef_manual)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_estimators_overwrite_params(name, estimator_orig):
X, y = make_blobs(random_state=0, n_samples=9)
# some want non-negative input
X -= X.min()
estimator = clone(estimator_orig)
y = multioutput_estimator_convert_y_2d(estimator, y)
set_random_state(estimator)
# Make a physical copy of the original estimator parameters before fitting.
params = estimator.get_params()
original_params = deepcopy(params)
# Fit the model
estimator.fit(X, y)
# Compare the state of the model parameters with the original parameters
new_params = estimator.get_params()
for param_name, original_value in original_params.items():
new_value = new_params[param_name]
# We should never change or mutate the internal state of input
# parameters by default. To check this we use the joblib.hash function
# that introspects recursively any subobjects to compute a checksum.
# The only exception to this rule of immutable constructor parameters
# is possible RandomState instance but in this check we explicitly
# fixed the random_state params recursively to be integer seeds.
assert_equal(hash(new_value), hash(original_value),
"Estimator %s should not change or mutate "
" the parameter %s from %s to %s during fit."
% (name, param_name, original_value, new_value))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_no_fit_attributes_set_in_init(name, Estimator):
"""Check that Estimator.__init__ doesn't set trailing-_ attributes."""
# this check works on classes, not instances
estimator = Estimator()
for attr in dir(estimator):
if attr.endswith("_") and not attr.startswith("__"):
# This check is for properties, they can be listed in dir
# while at the same time have hasattr return False as long
# as the property getter raises an AttributeError
assert_false(
hasattr(estimator, attr),
"By convention, attributes ending with '_' are "
'estimated from data in scikit-learn. Consequently they '
'should not be initialized in the constructor of an '
'estimator but in the fit method. Attribute {!r} '
'was found in estimator {}'.format(attr, name))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_sparsify_coefficients(name, estimator_orig):
X = np.array([[-2, -1], [-1, -1], [-1, -2], [1, 1], [1, 2], [2, 1],
[-1, -2], [2, 2], [-2, -2]])
y = [1, 1, 1, 2, 2, 2, 3, 3, 3]
est = clone(estimator_orig)
est.fit(X, y)
pred_orig = est.predict(X)
# test sparsify with dense inputs
est.sparsify()
assert_true(sparse.issparse(est.coef_))
pred = est.predict(X)
assert_array_equal(pred, pred_orig)
# pickle and unpickle with sparse coef_
est = pickle.loads(pickle.dumps(est))
assert_true(sparse.issparse(est.coef_))
pred = est.predict(X)
assert_array_equal(pred, pred_orig)
@ignore_warnings(category=DeprecationWarning)
def check_classifier_data_not_an_array(name, estimator_orig):
X = np.array([[3, 0], [0, 1], [0, 2], [1, 1], [1, 2], [2, 1]])
y = [1, 1, 1, 2, 2, 2]
y = multioutput_estimator_convert_y_2d(estimator_orig, y)
check_estimators_data_not_an_array(name, estimator_orig, X, y)
@ignore_warnings(category=DeprecationWarning)
def check_regressor_data_not_an_array(name, estimator_orig):
X, y = _boston_subset(n_samples=50)
y = multioutput_estimator_convert_y_2d(estimator_orig, y)
check_estimators_data_not_an_array(name, estimator_orig, X, y)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_estimators_data_not_an_array(name, estimator_orig, X, y):
if name in CROSS_DECOMPOSITION:
raise SkipTest
# separate estimators to control random seeds
estimator_1 = clone(estimator_orig)
estimator_2 = clone(estimator_orig)
set_random_state(estimator_1)
set_random_state(estimator_2)
y_ = NotAnArray(np.asarray(y))
X_ = NotAnArray(np.asarray(X))
# fit
estimator_1.fit(X_, y_)
pred1 = estimator_1.predict(X_)
estimator_2.fit(X, y)
pred2 = estimator_2.predict(X)
assert_allclose(pred1, pred2, atol=1e-2, err_msg=name)
def check_parameters_default_constructible(name, Estimator):
# this check works on classes, not instances
classifier = LinearDiscriminantAnalysis()
# test default-constructibility
# get rid of deprecation warnings
with ignore_warnings(category=(DeprecationWarning, FutureWarning)):
if name in META_ESTIMATORS:
estimator = Estimator(classifier)
else:
estimator = Estimator()
# test cloning
clone(estimator)
# test __repr__
repr(estimator)
# test that set_params returns self
assert_true(estimator.set_params() is estimator)
# test if init does nothing but set parameters
# this is important for grid_search etc.
# We get the default parameters from init and then
# compare these against the actual values of the attributes.
# this comes from getattr. Gets rid of deprecation decorator.
init = getattr(estimator.__init__, 'deprecated_original',
estimator.__init__)
try:
def param_filter(p):
"""Identify hyper parameters of an estimator"""
return (p.name != 'self' and
p.kind != p.VAR_KEYWORD and
p.kind != p.VAR_POSITIONAL)
init_params = [p for p in signature(init).parameters.values()
if param_filter(p)]
except (TypeError, ValueError):
# init is not a python function.
# true for mixins
return
params = estimator.get_params()
if name in META_ESTIMATORS:
# they can need a non-default argument
init_params = init_params[1:]
for init_param in init_params:
assert_not_equal(init_param.default, init_param.empty,
"parameter %s for %s has no default value"
% (init_param.name, type(estimator).__name__))
assert_in(type(init_param.default),
[str, int, float, bool, tuple, type(None),
np.float64, types.FunctionType, Memory])
if init_param.name not in params.keys():
# deprecated parameter, not in get_params
assert_true(init_param.default is None)
continue
if (issubclass(Estimator, BaseSGD) and
init_param.name in ['tol', 'max_iter']):
# To remove in 0.21, when they get their future default values
continue
param_value = params[init_param.name]
if isinstance(param_value, np.ndarray):
assert_array_equal(param_value, init_param.default)
else:
assert_equal(param_value, init_param.default, init_param.name)
def multioutput_estimator_convert_y_2d(estimator, y):
# Estimators in mono_output_task_error raise ValueError if y is of 1-D
# Convert into a 2-D y for those estimators.
if "MultiTask" in estimator.__class__.__name__:
return np.reshape(y, (-1, 1))
return y
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_non_transformer_estimators_n_iter(name, estimator_orig):
# Test that estimators that are not transformers with a parameter
# max_iter, return the attribute of n_iter_ at least 1.
# These models are dependent on external solvers like
# libsvm and accessing the iter parameter is non-trivial.
not_run_check_n_iter = ['Ridge', 'SVR', 'NuSVR', 'NuSVC',
'RidgeClassifier', 'SVC', 'RandomizedLasso',
'LogisticRegressionCV', 'LinearSVC',
'LogisticRegression']
# Tested in test_transformer_n_iter
not_run_check_n_iter += CROSS_DECOMPOSITION
if name in not_run_check_n_iter:
return
# LassoLars stops early for the default alpha=1.0 the iris dataset.
if name == 'LassoLars':
estimator = clone(estimator_orig).set_params(alpha=0.)
else:
estimator = clone(estimator_orig)
if hasattr(estimator, 'max_iter'):
iris = load_iris()
X, y_ = iris.data, iris.target
y_ = multioutput_estimator_convert_y_2d(estimator, y_)
set_random_state(estimator, 0)
if name == 'AffinityPropagation':
estimator.fit(X)
else:
estimator.fit(X, y_)
# HuberRegressor depends on scipy.optimize.fmin_l_bfgs_b
# which doesn't return a n_iter for old versions of SciPy.
if not (name == 'HuberRegressor' and estimator.n_iter_ is None):
assert_greater_equal(estimator.n_iter_, 1)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_transformer_n_iter(name, estimator_orig):
# Test that transformers with a parameter max_iter, return the
# attribute of n_iter_ at least 1.
estimator = clone(estimator_orig)
if hasattr(estimator, "max_iter"):
if name in CROSS_DECOMPOSITION:
# Check using default data
X = [[0., 0., 1.], [1., 0., 0.], [2., 2., 2.], [2., 5., 4.]]
y_ = [[0.1, -0.2], [0.9, 1.1], [0.1, -0.5], [0.3, -0.2]]
else:
X, y_ = make_blobs(n_samples=30, centers=[[0, 0, 0], [1, 1, 1]],
random_state=0, n_features=2, cluster_std=0.1)
X -= X.min() - 0.1
set_random_state(estimator, 0)
estimator.fit(X, y_)
# These return a n_iter per component.
if name in CROSS_DECOMPOSITION:
for iter_ in estimator.n_iter_:
assert_greater_equal(iter_, 1)
else:
assert_greater_equal(estimator.n_iter_, 1)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_get_params_invariance(name, estimator_orig):
# Checks if get_params(deep=False) is a subset of get_params(deep=True)
class T(BaseEstimator):
"""Mock classifier
"""
def __init__(self):
pass
def fit(self, X, y):
return self
def transform(self, X):
return X
e = clone(estimator_orig)
shallow_params = e.get_params(deep=False)
deep_params = e.get_params(deep=True)
assert_true(all(item in deep_params.items() for item in
shallow_params.items()))
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_classifiers_regression_target(name, estimator_orig):
# Check if classifier throws an exception when fed regression targets
boston = load_boston()
X, y = boston.data, boston.target
e = clone(estimator_orig)
msg = 'Unknown label type: '
assert_raises_regex(ValueError, msg, e.fit, X, y)
@ignore_warnings(category=(DeprecationWarning, FutureWarning))
def check_decision_proba_consistency(name, estimator_orig):
# Check whether an estimator having both decision_function and
# predict_proba methods has outputs with perfect rank correlation.
centers = [(2, 2), (4, 4)]
X, y = make_blobs(n_samples=100, random_state=0, n_features=4,
centers=centers, cluster_std=1.0, shuffle=True)
X_test = np.random.randn(20, 2) + 4
estimator = clone(estimator_orig)
if (hasattr(estimator, "decision_function") and
hasattr(estimator, "predict_proba")):
estimator.fit(X, y)
a = estimator.predict_proba(X_test)[:, 1]
b = estimator.decision_function(X_test)
assert_array_equal(rankdata(a), rankdata(b))