def test_forest_apply_result_shape(data):
X, y = data
clf = SimilarityForestClassifier()
clf.fit(X, y)
apply_result = clf.apply(X)
assert apply_result.shape == (X.shape[0], clf.n_estimators)
def test_probability_values_forest(data):
X, y = data
clf = SimilarityForestClassifier()
clf.fit(X, y)
preds = clf.predict_proba(X)
assert_allclose(np.sum(preds, axis=1), np.ones(shape=y.shape))
def test_similarity_forest_classifier_output_array_shape(data):
X, y = data
clf = SimilarityForestClassifier()
clf.fit(X, y)
y_pred = clf.predict(X)
assert y_pred.shape == (X.shape[0], )
def test_setting_attributes_forest(data):
X, y = data
clf = SimilarityForestClassifier(random_state=42, n_directions=2)
clf.fit(X, y)
y_pred = clf.predict(X)
assert clf.random_state == 42
assert clf.n_directions == 2
def test_train_set_acc(data):
X, y = data
forest = SimilarityForestClassifier()
forest.fit(X, y)
# shouldn't be actually 1.0?
assert forest.score(X, y) > 0.8
tree = SimilarityTreeClassifier()
tree.fit(X, y)
assert tree.score(X, y) > 0.9
def test_similarity_forest_outliers_ranking_stability(data):
"""There should not be a situation when models predicts the same when there is no random_state set"""
X, y = data
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=42)
clf = SimilarityForestClassifier()
clf.fit(X_train, y_train)
'''rcorrelations = clf.outliers_rank_stability(X_test, plot=False)
def test_deterministic_predictions_forest():
X, y = make_blobs(n_samples=300, centers=[(0, 0), (1, 1)], random_state=42)
clf1 = SimilarityForestClassifier(random_state=42)
clf1.fit(X, y)
clf2 = SimilarityForestClassifier(random_state=42)
clf2.fit(X, y)
y_pred1 = clf1.predict(X)
y_pred2 = clf2.predict(X)
assert_array_equal(y_pred1, y_pred2)
def test_similarity_forest_classifier_prediction(data):
X, y = data
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=42)
clf = SimilarityForestClassifier()
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
assert y_pred.shape == (X_test.shape[0], )
assert accuracy_score(y_test, y_pred) > 0.9
def test_similarity_forest_wrongly_the_same_pred(data):
"""There should not be a situation when models predicts the same when there is no random_state set"""
X, y = data
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=42)
clf1 = SimilarityForestClassifier()
clf1.fit(X_train, y_train)
y_pred1 = clf1.predict_proba(X_test)
clf2 = SimilarityForestClassifier()
clf2.fit(X_train, y_train)
y_pred2 = clf2.predict_proba(X_test)
assert not np.array_equal(y_pred1, y_pred2)
def test_log_probabilities_forest(data):
X, y = data
clf = SimilarityForestClassifier()
clf.fit(X, y)
preds = clf.predict_proba(X)
log_preds = clf.predict_log_proba(X)
assert_allclose(log_preds, np.log(preds + 1e-10))
for d_idx, d in enumerate(get_datasets()):
X_train, X_test, y_train, y_test, dataset = d
if binary:
scorer = 'f1'
else:
scorer = 'f1_weighted'
# Find parameters
# RF
rf = GridSearchCV(RandomForestClassifier(), param_grid=rf_params, cv=5, scoring=scorer, refit=scorer, n_jobs=4)
rf.fit(X_train, y_train)
# SF
if multiclass_strategy:
ovr = OneVsRestClassifier(SimilarityForestClassifier())
sf = GridSearchCV(ovr, param_grid=ovr_sf_params, cv=n_folds, verbose=5, n_jobs=4, scoring='f1', refit='f1')
else:
sf = GridSearchCV(SimilarityForestClassifier(), param_grid=sf_params, cv=n_folds, verbose=5, n_jobs=4, scoring=scorer, refit=scorer)
sf.fit(X_train, y_train)
# Log std of CV scores for different parameters
rf_std_score = np.std(rf.cv_results_['mean_test_score'])
if use_neptune:
neptune.log_metric(f'{dataset} RF std cv score', rf_std_score)
sf_std_score = np.std(sf.cv_results_['mean_test_score'])
if use_neptune:
neptune.log_metric(f'{dataset} SF std cv score', sf_std_score)