def test_predict(self):
"""...Test PoissonRegression predict
"""
X_train, y_train, _, _ = self.get_train_data(n_samples=200,
n_features=12)
learner = PoissonRegression(random_state=32789, tol=1e-9)
learner.fit(X_train, y_train)
X_test, y_test, _, _ = self.get_train_data(n_samples=5, n_features=12)
y_pred = np.array([1., 5., 0., 5., 6.])
np.testing.assert_array_almost_equal(learner.predict(X_test), y_pred)
def test_loglik(self):
"""...Test PoissonRegression loglik function
"""
X_train, y_train, _, _ = self.get_train_data(n_samples=200,
n_features=12)
learner = PoissonRegression(random_state=32789, tol=1e-9)
learner.fit(X_train, y_train)
X_test, y_test, _, _ = self.get_train_data(n_samples=5, n_features=12)
np.testing.assert_array_almost_equal(learner.loglik(X_test, y_test),
1.8254,
decimal=4)
def test_decision_function(self):
"""...Test PoissonRegression decision function
"""
X_train, y_train, _, _ = self.get_train_data(n_samples=200,
n_features=12)
learner = PoissonRegression(random_state=32789, tol=1e-9)
learner.fit(X_train, y_train)
X_test, y_test, _, _ = self.get_train_data(n_samples=5, n_features=12)
y_pred = np.array([1.1448, 5.2194, 0.2624, 4.5525, 6.4168])
np.testing.assert_array_almost_equal(learner.decision_function(X_test),
y_pred,
decimal=4)
def test_PoissonRegression_fit(self):
"""...Test PoissonRegression fit with default parameters
"""
n_samples = 2000
n_features = 20
for fit_intercept in [False, True]:
X, y, weights0, intercept0 = self.get_train_data(
n_samples=n_samples,
n_features=n_features,
fit_intercept=fit_intercept)
learner = PoissonRegression(C=1e3,
verbose=False,
fit_intercept=fit_intercept,
solver='bfgs')
learner.fit(X, y)
err = np.linalg.norm(learner.weights - weights0) / n_features
self.assertLess(err, 1e-2)
if fit_intercept:
self.assertLess(np.abs(learner.intercept - intercept0), 1e-1)
def test_PoissonRegression_run(self):
"""...Test PoissonRegression runs with different solvers and penalties
"""
n_samples = 200
n_features = 10
for fit_intercept in [False, True]:
X, y, weights0, intercept0 = self.get_train_data(
n_samples=n_samples,
n_features=n_features,
fit_intercept=fit_intercept)
for solver, penalty in product(PoissonRegression._solvers,
PoissonRegression._penalties):
if solver == 'bfgs' and (penalty not in ['zero', 'l2']):
continue
if penalty == 'binarsity':
learner = PoissonRegression(verbose=False,
fit_intercept=fit_intercept,
solver=solver,
penalty=penalty,
max_iter=1,
step=1e-5,
blocks_start=[0],
blocks_length=[1])
else:
learner = PoissonRegression(verbose=False,
fit_intercept=fit_intercept,
solver=solver,
penalty=penalty,
max_iter=1,
step=1e-5)
learner.fit(X, y)
self.assertTrue(np.isfinite(learner.weights).all())
if fit_intercept:
self.assertTrue(np.isfinite(learner.intercept))