def test_make_multilabel_classification_return_indicator():
for allow_unlabeled, min_length in zip((True, False), (0, 1)):
X, Y = make_multilabel_classification(n_samples=25,
n_features=20,
n_classes=3,
random_state=0,
allow_unlabeled=allow_unlabeled)
assert X.shape == (25, 20), "X shape mismatch"
assert Y.shape == (25, 3), "Y shape mismatch"
assert np.all(np.sum(Y, axis=0) > min_length)
# Also test return_distributions and return_indicator with True
X2, Y2, p_c, p_w_c = make_multilabel_classification(
n_samples=25,
n_features=20,
n_classes=3,
random_state=0,
allow_unlabeled=allow_unlabeled,
return_distributions=True)
assert_array_almost_equal(X, X2)
assert_array_equal(Y, Y2)
assert p_c.shape == (3, )
assert_almost_equal(p_c.sum(), 1)
assert p_w_c.shape == (20, 3)
assert_almost_equal(p_w_c.sum(axis=0), [1] * 3)
def test_multilabel_classification():
# Test that multi-label classification works as expected.
# test fit method
X, y = make_multilabel_classification(n_samples=50,
random_state=0,
return_indicator=True)
mlp = MLPClassifier(solver='lbfgs',
hidden_layer_sizes=50,
alpha=1e-5,
max_iter=150,
random_state=0,
activation='logistic',
learning_rate_init=0.2)
mlp.fit(X, y)
assert mlp.score(X, y) > 0.97
# test partial fit method
mlp = MLPClassifier(solver='sgd',
hidden_layer_sizes=50,
max_iter=150,
random_state=0,
activation='logistic',
alpha=1e-5,
learning_rate_init=0.2)
for i in range(100):
mlp.partial_fit(X, y, classes=[0, 1, 2, 3, 4])
assert mlp.score(X, y) > 0.9
# Make sure early stopping still work now that spliting is stratified by
# default (it is disabled for multilabel classification)
mlp = MLPClassifier(early_stopping=True)
mlp.fit(X, y).predict(X)
def setup_module():
# Create some memory mapped data
global X_mm, y_mm, y_ml_mm, TEMP_FOLDER, ESTIMATORS
TEMP_FOLDER = tempfile.mkdtemp(prefix='mrex_test_score_objects_')
X, y = make_classification(n_samples=30, n_features=5, random_state=0)
_, y_ml = make_multilabel_classification(n_samples=X.shape[0],
random_state=0)
filename = os.path.join(TEMP_FOLDER, 'test_data.pkl')
joblib.dump((X, y, y_ml), filename)
X_mm, y_mm, y_ml_mm = joblib.load(filename, mmap_mode='r')
ESTIMATORS = _make_estimators(X_mm, y_mm, y_ml_mm)
def test_make_multilabel_classification_return_indicator_sparse():
for allow_unlabeled, min_length in zip((True, False), (0, 1)):
X, Y = make_multilabel_classification(n_samples=25,
n_features=20,
n_classes=3,
random_state=0,
return_indicator='sparse',
allow_unlabeled=allow_unlabeled)
assert X.shape == (25, 20), "X shape mismatch"
assert Y.shape == (25, 3), "Y shape mismatch"
assert sp.issparse(Y)
def test_make_multilabel_classification_return_sequences():
for allow_unlabeled, min_length in zip((True, False), (0, 1)):
X, Y = make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=3,
random_state=0,
return_indicator=False,
allow_unlabeled=allow_unlabeled)
assert X.shape == (100, 20), "X shape mismatch"
if not allow_unlabeled:
assert max([max(y) for y in Y]) == 2
assert min([len(y) for y in Y]) == min_length
assert max([len(y) for y in Y]) <= 3
def test_multilabel():
"""Check if error is raised for multilabel classification."""
X, y = make_multilabel_classification(n_classes=2, n_labels=1,
allow_unlabeled=False,
random_state=123)
clf = OneVsRestClassifier(SVC(kernel='linear'))
eclf = VotingClassifier(estimators=[('ovr', clf)], voting='hard')
try:
eclf.fit(X, y)
except NotImplementedError:
return
def test_ovr_multilabel_decision_function():
X, Y = datasets.make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=5,
n_labels=3,
length=50,
allow_unlabeled=True,
random_state=0)
X_train, Y_train = X[:80], Y[:80]
X_test = X[80:]
clf = OneVsRestClassifier(svm.SVC()).fit(X_train, Y_train)
assert_array_equal((clf.decision_function(X_test) > 0).astype(int),
clf.predict(X_test))
def test_scorer_sample_weight():
# Test that scorers support sample_weight or raise sensible errors
# Unlike the metrics invariance test, in the scorer case it's harder
# to ensure that, on the classifier output, weighted and unweighted
# scores really should be unequal.
X, y = make_classification(random_state=0)
_, y_ml = make_multilabel_classification(n_samples=X.shape[0],
random_state=0)
split = train_test_split(X, y, y_ml, random_state=0)
X_train, X_test, y_train, y_test, y_ml_train, y_ml_test = split
sample_weight = np.ones_like(y_test)
sample_weight[:10] = 0
# get sensible estimators for each metric
estimator = _make_estimators(X_train, y_train, y_ml_train)
for name, scorer in SCORERS.items():
if name in MULTILABEL_ONLY_SCORERS:
target = y_ml_test
else:
target = y_test
if name in REQUIRE_POSITIVE_Y_SCORERS:
target = _require_positive_y(target)
try:
weighted = scorer(estimator[name],
X_test,
target,
sample_weight=sample_weight)
ignored = scorer(estimator[name], X_test[10:], target[10:])
unweighted = scorer(estimator[name], X_test, target)
assert weighted != unweighted, (
"scorer {0} behaves identically when "
"called with sample weights: {1} vs "
"{2}".format(name, weighted, unweighted))
assert_almost_equal(weighted,
ignored,
err_msg="scorer {0} behaves differently when "
"ignoring samples and setting sample_weight to"
" 0: {1} vs {2}".format(
name, weighted, ignored))
except TypeError as e:
assert "sample_weight" in str(e), (
"scorer {0} raises unhelpful exception when called "
"with sample weights: {1}".format(name, str(e)))
def test_ovr_fit_predict_sparse():
for sparse in [
sp.csr_matrix, sp.csc_matrix, sp.coo_matrix, sp.dok_matrix,
sp.lil_matrix
]:
base_clf = MultinomialNB(alpha=1)
X, Y = datasets.make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=5,
n_labels=3,
length=50,
allow_unlabeled=True,
random_state=0)
X_train, Y_train = X[:80], Y[:80]
X_test = X[80:]
clf = OneVsRestClassifier(base_clf).fit(X_train, Y_train)
Y_pred = clf.predict(X_test)
clf_sprs = OneVsRestClassifier(base_clf).fit(X_train, sparse(Y_train))
Y_pred_sprs = clf_sprs.predict(X_test)
assert clf.multilabel_
assert sp.issparse(Y_pred_sprs)
assert_array_equal(Y_pred_sprs.toarray(), Y_pred)
# Test predict_proba
Y_proba = clf_sprs.predict_proba(X_test)
# predict assigns a label if the probability that the
# sample has the label is greater than 0.5.
pred = Y_proba > .5
assert_array_equal(pred, Y_pred_sprs.toarray())
# Test decision_function
clf = svm.SVC()
clf_sprs = OneVsRestClassifier(clf).fit(X_train, sparse(Y_train))
dec_pred = (clf_sprs.decision_function(X_test) > 0).astype(int)
assert_array_equal(dec_pred, clf_sprs.predict(X_test).toarray())
def test_thresholded_scorers_multilabel_indicator_data():
# Test that the scorer work with multilabel-indicator format
# for multilabel and multi-output multi-class classifier
X, y = make_multilabel_classification(allow_unlabeled=False,
random_state=0)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
# Multi-output multi-class predict_proba
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
y_proba = clf.predict_proba(X_test)
score1 = get_scorer('roc_auc')(clf, X_test, y_test)
score2 = roc_auc_score(y_test, np.vstack([p[:, -1] for p in y_proba]).T)
assert_almost_equal(score1, score2)
# Multi-output multi-class decision_function
# TODO Is there any yet?
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
clf._predict_proba = clf.predict_proba
clf.predict_proba = None
clf.decision_function = lambda X: [p[:, 1] for p in clf._predict_proba(X)]
y_proba = clf.decision_function(X_test)
score1 = get_scorer('roc_auc')(clf, X_test, y_test)
score2 = roc_auc_score(y_test, np.vstack([p for p in y_proba]).T)
assert_almost_equal(score1, score2)
# Multilabel predict_proba
clf = OneVsRestClassifier(DecisionTreeClassifier())
clf.fit(X_train, y_train)
score1 = get_scorer('roc_auc')(clf, X_test, y_test)
score2 = roc_auc_score(y_test, clf.predict_proba(X_test))
assert_almost_equal(score1, score2)
# Multilabel decision function
clf = OneVsRestClassifier(LinearSVC(random_state=0))
clf.fit(X_train, y_train)
score1 = get_scorer('roc_auc')(clf, X_test, y_test)
score2 = roc_auc_score(y_test, clf.decision_function(X_test))
assert_almost_equal(score1, score2)
def test_ovr_multilabel_predict_proba():
base_clf = MultinomialNB(alpha=1)
for au in (False, True):
X, Y = datasets.make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=5,
n_labels=3,
length=50,
allow_unlabeled=au,
random_state=0)
X_train, Y_train = X[:80], Y[:80]
X_test = X[80:]
clf = OneVsRestClassifier(base_clf).fit(X_train, Y_train)
# Decision function only estimator.
decision_only = OneVsRestClassifier(svm.SVR()).fit(X_train, Y_train)
assert not hasattr(decision_only, 'predict_proba')
# Estimator with predict_proba disabled, depending on parameters.
decision_only = OneVsRestClassifier(svm.SVC(probability=False))
assert not hasattr(decision_only, 'predict_proba')
decision_only.fit(X_train, Y_train)
assert not hasattr(decision_only, 'predict_proba')
assert hasattr(decision_only, 'decision_function')
# Estimator which can get predict_proba enabled after fitting
gs = GridSearchCV(svm.SVC(probability=False),
param_grid={'probability': [True]})
proba_after_fit = OneVsRestClassifier(gs)
assert not hasattr(proba_after_fit, 'predict_proba')
proba_after_fit.fit(X_train, Y_train)
assert hasattr(proba_after_fit, 'predict_proba')
Y_pred = clf.predict(X_test)
Y_proba = clf.predict_proba(X_test)
# predict assigns a label if the probability that the
# sample has the label is greater than 0.5.
pred = Y_proba > .5
assert_array_equal(pred, Y_pred)
def test_ovr_multilabel_dataset():
base_clf = MultinomialNB(alpha=1)
for au, prec, recall in zip((True, False), (0.51, 0.66), (0.51, 0.80)):
X, Y = datasets.make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=5,
n_labels=2,
length=50,
allow_unlabeled=au,
random_state=0)
X_train, Y_train = X[:80], Y[:80]
X_test, Y_test = X[80:], Y[80:]
clf = OneVsRestClassifier(base_clf).fit(X_train, Y_train)
Y_pred = clf.predict(X_test)
assert clf.multilabel_
assert_almost_equal(precision_score(Y_test, Y_pred, average="micro"),
prec,
decimal=2)
assert_almost_equal(recall_score(Y_test, Y_pred, average="micro"),
recall,
decimal=2)
def test_predict_proba_multilabel():
# Test that predict_proba works as expected for multilabel.
# Multilabel should not use softmax which makes probabilities sum to 1
X, Y = make_multilabel_classification(n_samples=50,
random_state=0,
return_indicator=True)
n_samples, n_classes = Y.shape
clf = MLPClassifier(solver='lbfgs', hidden_layer_sizes=30, random_state=0)
clf.fit(X, Y)
y_proba = clf.predict_proba(X)
assert y_proba.shape == (n_samples, n_classes)
assert_array_equal(y_proba > 0.5, Y)
y_log_proba = clf.predict_log_proba(X)
proba_max = y_proba.argmax(axis=1)
proba_log_max = y_log_proba.argmax(axis=1)
assert (y_proba.sum(1) - 1).dot(y_proba.sum(1) - 1) > 1e-10
assert_array_equal(proba_max, proba_log_max)
assert_array_equal(y_log_proba, np.log(y_proba))
def test_ridge_classifier_no_support_multilabel():
X, y = make_multilabel_classification(n_samples=10, random_state=0)
assert_raises(ValueError, RidgeClassifier().fit, X, y)
def test_sparse_classification():
# Check classification with sparse input.
class CustomSVC(SVC):
"""SVC variant that records the nature of the training set."""
def fit(self, X, y, sample_weight=None):
"""Modification on fit caries data type for later verification."""
super().fit(X, y, sample_weight=sample_weight)
self.data_type_ = type(X)
return self
X, y = datasets.make_multilabel_classification(n_classes=1,
n_samples=15,
n_features=5,
random_state=42)
# Flatten y to a 1d array
y = np.ravel(y)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
for sparse_format in [
csc_matrix, csr_matrix, lil_matrix, coo_matrix, dok_matrix
]:
X_train_sparse = sparse_format(X_train)
X_test_sparse = sparse_format(X_test)
# Trained on sparse format
sparse_classifier = AdaBoostClassifier(
base_estimator=CustomSVC(probability=True),
random_state=1,
algorithm="SAMME").fit(X_train_sparse, y_train)
# Trained on dense format
dense_classifier = AdaBoostClassifier(
base_estimator=CustomSVC(probability=True),
random_state=1,
algorithm="SAMME").fit(X_train, y_train)
# predict
sparse_results = sparse_classifier.predict(X_test_sparse)
dense_results = dense_classifier.predict(X_test)
assert_array_equal(sparse_results, dense_results)
# decision_function
sparse_results = sparse_classifier.decision_function(X_test_sparse)
dense_results = dense_classifier.decision_function(X_test)
assert_array_almost_equal(sparse_results, dense_results)
# predict_log_proba
sparse_results = sparse_classifier.predict_log_proba(X_test_sparse)
dense_results = dense_classifier.predict_log_proba(X_test)
assert_array_almost_equal(sparse_results, dense_results)
# predict_proba
sparse_results = sparse_classifier.predict_proba(X_test_sparse)
dense_results = dense_classifier.predict_proba(X_test)
assert_array_almost_equal(sparse_results, dense_results)
# score
sparse_results = sparse_classifier.score(X_test_sparse, y_test)
dense_results = dense_classifier.score(X_test, y_test)
assert_array_almost_equal(sparse_results, dense_results)
# staged_decision_function
sparse_results = sparse_classifier.staged_decision_function(
X_test_sparse)
dense_results = dense_classifier.staged_decision_function(X_test)
for sprase_res, dense_res in zip(sparse_results, dense_results):
assert_array_almost_equal(sprase_res, dense_res)
# staged_predict
sparse_results = sparse_classifier.staged_predict(X_test_sparse)
dense_results = dense_classifier.staged_predict(X_test)
for sprase_res, dense_res in zip(sparse_results, dense_results):
assert_array_equal(sprase_res, dense_res)
# staged_predict_proba
sparse_results = sparse_classifier.staged_predict_proba(X_test_sparse)
dense_results = dense_classifier.staged_predict_proba(X_test)
for sprase_res, dense_res in zip(sparse_results, dense_results):
assert_array_almost_equal(sprase_res, dense_res)
# staged_score
sparse_results = sparse_classifier.staged_score(X_test_sparse, y_test)
dense_results = dense_classifier.staged_score(X_test, y_test)
for sprase_res, dense_res in zip(sparse_results, dense_results):
assert_array_equal(sprase_res, dense_res)
# Verify sparsity of data is maintained during training
types = [i.data_type_ for i in sparse_classifier.estimators_]
assert all([(t == csc_matrix or t == csr_matrix) for t in types])
def test_sparse_input(name, sparse_matrix):
X, y = datasets.make_multilabel_classification(random_state=0,
n_samples=50)
check_sparse_input(name, X, sparse_matrix(X), y)
def test_random_hasher_sparse_data():
X, y = datasets.make_multilabel_classification(random_state=0)
hasher = RandomTreesEmbedding(n_estimators=30, random_state=1)
X_transformed = hasher.fit_transform(X)
X_transformed_sparse = hasher.fit_transform(csc_matrix(X))
assert_array_equal(X_transformed_sparse.toarray(), X_transformed.toarray())
def benchmark(metrics=tuple(v for k, v in sorted(METRICS.items())),
formats=tuple(v for k, v in sorted(FORMATS.items())),
samples=1000,
classes=4,
density=.2,
n_times=5):
"""Times metric calculations for a number of inputs
Parameters
----------
metrics : array-like of callables (1d or 0d)
The metric functions to time.
formats : array-like of callables (1d or 0d)
These may transform a dense indicator matrix into multilabel
representation.
samples : array-like of ints (1d or 0d)
The number of samples to generate as input.
classes : array-like of ints (1d or 0d)
The number of classes in the input.
density : array-like of ints (1d or 0d)
The density of positive labels in the input.
n_times : int
Time calling the metric n_times times.
Returns
-------
array of floats shaped like (metrics, formats, samples, classes, density)
Time in seconds.
"""
metrics = np.atleast_1d(metrics)
samples = np.atleast_1d(samples)
classes = np.atleast_1d(classes)
density = np.atleast_1d(density)
formats = np.atleast_1d(formats)
out = np.zeros(
(len(metrics), len(formats), len(samples), len(classes), len(density)),
dtype=float)
it = itertools.product(samples, classes, density)
for i, (s, c, d) in enumerate(it):
_, y_true = make_multilabel_classification(n_samples=s,
n_features=1,
n_classes=c,
n_labels=d * c,
random_state=42)
_, y_pred = make_multilabel_classification(n_samples=s,
n_features=1,
n_classes=c,
n_labels=d * c,
random_state=84)
for j, f in enumerate(formats):
f_true = f(y_true)
f_pred = f(y_pred)
for k, metric in enumerate(metrics):
t = timeit(partial(metric, f_true, f_pred), number=n_times)
out[k, j].flat[i] = t
return out
'Boundary\nfor class 2')
plt.xticks(())
plt.yticks(())
plt.xlim(min_x - .5 * max_x, max_x + .5 * max_x)
plt.ylim(min_y - .5 * max_y, max_y + .5 * max_y)
if subplot == 2:
plt.xlabel('First principal component')
plt.ylabel('Second principal component')
plt.legend(loc="upper left")
plt.figure(figsize=(8, 6))
X, Y = make_multilabel_classification(n_classes=2,
n_labels=1,
allow_unlabeled=True,
random_state=1)
plot_subfigure(X, Y, 1, "With unlabeled samples + CCA", "cca")
plot_subfigure(X, Y, 2, "With unlabeled samples + PCA", "pca")
X, Y = make_multilabel_classification(n_classes=2,
n_labels=1,
allow_unlabeled=False,
random_state=1)
plot_subfigure(X, Y, 3, "Without unlabeled samples + CCA", "cca")
plot_subfigure(X, Y, 4, "Without unlabeled samples + PCA", "pca")
plt.subplots_adjust(.04, .02, .97, .94, .09, .2)
plt.show()