def test_predict_proba(example_static_selection):
X, y, pool = example_static_selection
expected = np.tile([0.52, 0.48], (y.size, 1))
static_selection_test = StaticSelection(pool, 0.5)
static_selection_test.fit(X, y)
proba = static_selection_test.predict_proba(X)
assert np.allclose(proba, expected)
def test_static_selection():
pool_classifiers, X_dsel, y_dsel, X_test, y_test = setup_classifiers()
static_selection = StaticSelection(pool_classifiers)
static_selection.fit(X_dsel, y_dsel)
assert np.isclose(static_selection.score(X_test, y_test),
0.90606060606060601)
def test_predict_diff(example_static_selection):
X, y, pool = example_static_selection
static_selection_test = StaticSelection(pool, 0.75)
static_selection_test.fit(X, y)
predicted_labels = static_selection_test.predict(X)
assert np.equal(predicted_labels, 1).all()
def test_predict(example_static_selection, create_pool_classifiers):
X, y, _ = example_static_selection
static_selection_test = StaticSelection(create_pool_classifiers * 10, 0.25)
static_selection_test.fit(X, y)
predicted_labels = static_selection_test.predict(X)
assert np.equal(predicted_labels, 0).all()
def test_predict():
X = X_dsel_ex1
y = y_dsel_ex1
pool_classifiers = create_pool_classifiers_score(
1, 25, 0.5) + create_pool_classifiers_score(0, 25, 1.0)
static_selection_test = StaticSelection(pool_classifiers, 0.25)
static_selection_test.fit(X, y)
predicted_labels = static_selection_test.predict(X)
assert np.equal(predicted_labels, 0).all()
def test_fit(example_static_selection):
X, y, pool = example_static_selection
static_selection_test = StaticSelection(pool, 0.5)
static_selection_test.fit(X, y)
assert static_selection_test.n_classifiers_ensemble_ == 50
assert static_selection_test.n_classifiers_ensemble_ == len(
static_selection_test.clf_indices_)
assert np.array_equal(np.sort(static_selection_test.clf_indices_),
list(range(50, 100)))
def test_label_encoder_base_ensemble():
from sklearn.ensemble import RandomForestClassifier
X, y = make_classification()
y[y == 1] = 2
y = y.astype(np.float)
pool = RandomForestClassifier().fit(X, y)
ss = StaticSelection(pool)
ss.fit(X, y)
pred = ss.predict(X)
assert np.isin(ss.classes_, pred).all()
def test_different_scorer():
X, y = make_classification(n_samples=100, random_state=42)
X_val, y_val = make_classification(n_samples=25, random_state=123)
pool = AdaBoostClassifier(n_estimators=10).fit(X, y)
performances = []
for clf in pool:
preds = clf.predict_proba(X_val)
performances.append(log_loss(y_val.ravel(), preds[:, -1]))
id_best = np.argsort(performances)
ss = StaticSelection(pool_classifiers=pool, scoring='neg_log_loss')
ss.fit(X_val, y_val)
assert (id_best[:ss.n_classifiers_ensemble_] == ss.clf_indices_).all()
def test_fit():
X = X_dsel_ex1
y = y_dsel_ex1
pool_classifiers = create_pool_classifiers_score(
1, 50, 0.5) + create_pool_classifiers_score(1, 50, 1.0)
static_selection_test = StaticSelection(pool_classifiers, 0.5)
static_selection_test.fit(X, y)
assert static_selection_test.n_classifiers_ensemble == 50
assert static_selection_test.n_classifiers_ensemble == len(
static_selection_test.clf_indices)
assert np.array_equal(np.sort(static_selection_test.clf_indices),
list(range(50, 100)))
def test_invalid_pct2():
with pytest.raises(ValueError):
test = StaticSelection(pct_classifiers=1.2)
test.fit(np.random.rand(10, 2), np.ones(10))
def test_invalid_pct():
with pytest.raises(TypeError):
test = StaticSelection(pct_classifiers='something')
test.fit(np.random.rand(10, 2), np.ones(10))