def test_approx_cross_validation_OutSamplerTransformer_regressor(multi_output):
np.random.seed(123)
X = np.random.randn(100, 10)
if multi_output:
y = np.random.randn(100, 2)
else:
y = np.random.randn(100)
model = OutSamplerTransformer(RandomForestRegressor(n_estimators=10,
random_state=123),
cv=10)
cv_res, yhat = model.approx_cross_validation(X,
y,
cv=10,
method="transform",
no_scoring=True)
assert cv_res is None
assert yhat.ndim == 2
if multi_output:
assert yhat.shape == y.shape
else:
assert yhat.shape == (y.shape[0], 1)
with pytest.raises(NotFittedError):
model.transform(X)
cv = KFold(n_splits=10, shuffle=True, random_state=123)
model = OutSamplerTransformer(DummyRegressor(), cv=cv)
yhat1 = model.fit_transform(X, y)
cv_res, yhat2 = model.approx_cross_validation(X,
y,
cv=cv,
method="transform",
no_scoring=True,
return_predict=True)
# Approx cross val and fit transform should return the same thing here
assert np.abs((yhat1 - yhat2).flatten()).max() <= 10**(-5)
if multi_output:
yhat3 = np.zeros(y.shape)
else:
yhat3 = np.zeros((y.shape[0], 1))
for train, test in cv.split(X, y):
model = DummyRegressor()
model.fit(X[train, :], y[train])
if multi_output:
yhat3[test, :] = model.predict(X[test, :])
else:
yhat3[test, 0] = model.predict(X[test, :])
assert np.abs((yhat1 - yhat3).flatten()).max() <= 10**(-5)
assert np.abs((yhat1 - yhat2).flatten()).max() <= 10**(-5)
def test_OutSampleTransformer_classifier_unbalanced():
np.random.seed(123)
X = np.random.randn(100, 2)
y = np.array(["AA"] * 33 + ["BB"] * 33 + ["CC"] * 33 + ["DD"])
model = OutSamplerTransformer(
RandomForestClassifier(n_estimators=10, random_state=123))
p3 = model.fit_transform(X, y)
assert (p3.max(axis=1) > 0).all()
def test_approx_cross_validation_OutSamplerTransformer_classifier():
np.random.seed(123)
X = np.random.randn(100, 10)
y = 1 * (np.random.randn(100) > 0)
model = OutSamplerTransformer(RandomForestClassifier(random_state=123),
cv=10)
cv_res, yhat = model.approx_cross_validation(X,
y,
cv=10,
method="transform",
no_scoring=True)
assert cv_res is None
assert yhat.ndim == 2
assert yhat.shape == (y.shape[0], 1)
with pytest.raises(NotFittedError):
model.transform(X)
with pytest.raises(NotFittedError):
model.model.predict(X)
cv = KFold(n_splits=10, shuffle=True, random_state=123)
model = OutSamplerTransformer(LogisticRegression(C=1, random_state=123),
cv=cv)
yhat1 = model.fit_transform(X, y)
model = OutSamplerTransformer(LogisticRegression(C=1, random_state=123),
cv=cv)
cv_res, yhat2 = model.approx_cross_validation(X,
y,
cv=cv,
method="transform",
no_scoring=True,
return_predict=True)
# Approx cross val and fit transform should return the same thing here
assert np.abs((yhat1 - yhat2).flatten()).max() <= 10**(-5)
yhat3 = np.zeros((y.shape[0], 1))
for train, test in cv.split(X, y):
model = LogisticRegression()
model.fit(X[train, :], y[train])
yhat3[test, 0] = model.predict_proba(X[test, :])[:, 1]
assert np.abs((yhat1 - yhat3).flatten()).max() <= 10**(-5)
assert np.abs((yhat1 - yhat2).flatten()).max() <= 10**(-5)
def test_OutSamplerTransformer_regressor_fit_transform():
np.random.seed(123)
X = np.random.randn(100, 10)
y = np.random.randn(100)
cv = KFold(n_splits=10, shuffle=True, random_state=123)
model = OutSamplerTransformer(DummyRegressor(), cv=cv)
model.fit(X, y)
y1 = model.transform(X)
model = OutSamplerTransformer(DummyRegressor(), cv=cv)
y2 = model.fit_transform(X, y)
assert np.abs(y1 -
y2).flatten().max() >= 0.01 # vector should be different
def test_OutSamplerTransformer_classifier_fit_transform():
X = np.random.randn(100, 10)
y = 1 * (np.random.randn(100) > 0)
cv = KFold(n_splits=10, shuffle=True, random_state=123)
model = OutSamplerTransformer(LogisticRegression(C=1, random_state=123),
cv=cv)
model.fit(X, y)
y1 = model.transform(X)
model = OutSamplerTransformer(LogisticRegression(C=1, random_state=123),
cv=cv)
y2 = model.fit_transform(X, y)
assert np.abs(y1 -
y2).flatten().max() >= 0.01 # vector should be different
def test_OutSamplerTransformer_classifier_fit_transform(multi_output):
X = np.random.randn(100, 10)
if multi_output:
y = 1 * (np.random.randn(100, 2) > 0)
else:
y = 1 * (np.random.randn(100) > 0)
cv = KFold(n_splits=10, shuffle=True, random_state=123)
model = OutSamplerTransformer(RandomForestClassifier(n_estimators=10,
random_state=123),
cv=cv)
model.fit(X, y)
y1 = model.transform(X)
model = OutSamplerTransformer(RandomForestClassifier(n_estimators=10,
random_state=123),
cv=cv)
y2 = model.fit_transform(X, y)
assert np.abs(y1 -
y2).flatten().max() >= 0.01 # vector should be different