def test_ols(self):
self.model = OLS(
solver='pinv', fit_intercept=True, est_stderr=True,
reg_method=None, alpha=0, l1_ratio=0, tol=1e-4, max_iter=100,
coef=None, stderr=None, dispersion=None)
self.model.fit(self.data_longley.exog, self.data_longley.endog)
# coefficient
self.assertEqual(self.model.coef.shape, (1, 7))
np.testing.assert_array_almost_equal(
self.model.coef,
np.array([-3482258.63459582, 15.0618722713733, -0.358191792925910E-01,
-2.02022980381683, -1.03322686717359, -0.511041056535807E-01,
1829.15146461355]).reshape(1, -1),
decimal=3)
# std.err of coefficient (calibrated by df_resid)
self.assertEqual(self.model.stderr.shape, (1, 7))
np.testing.assert_array_almost_equal(
old_div(self.model.stderr, np.sqrt(old_div(9., self.data_longley.exog.shape[0]))),
np.array([890420.383607373, 84.9149257747669, 0.03349,
0.488399681651699, 0.214274163161675, 0.226073200069370,
455.478499142212]).reshape(1, -1),
decimal=2)
# scale
self.assertEqual(self.model.dispersion.shape, (1, 1))
np.testing.assert_array_almost_equal(
old_div(self.model.dispersion, (old_div(9., self.data_longley.exog.shape[0]))),
np.array([[92936.0061673238]]),
decimal=3)
# predict
np.testing.assert_array_almost_equal(
self.data_longley.endog.reshape(-1, 1) - self.model.predict(self.data_longley.exog),
np.array([267.34003, -94.01394, 46.28717, -410.11462,
309.71459, -249.31122, -164.04896, -13.18036, 14.30477, 455.39409,
-17.26893, -39.05504, -155.54997, -85.67131, 341.93151,
-206.75783]).reshape(-1, 1),
decimal=3)
# loglike/_per_sample
self.assertAlmostEqual(
self.model.loglike(self.data_longley.exog, self.data_longley.endog),
-109.61743480849013,
places=3)
# to_json
json_dict = self.model.to_json('./tests/linear_models/OLS/UnivariateOLS/')
self.assertEqual(json_dict['properties']['solver'], 'pinv')
# from_json
self.model_from_json = OLS.from_json(json_dict)
np.testing.assert_array_almost_equal(
self.model.coef,
self.model_from_json.coef,
decimal=3)
np.testing.assert_array_almost_equal(
self.model.stderr,
self.model_from_json.stderr,
decimal=3)
self.assertEqual(
self.model.dispersion,
self.model_from_json.dispersion)
def test_ols(self):
self.model = OLS(
solver='pinv', fit_intercept=True, est_stderr=True,
reg_method=None, alpha=0, l1_ratio=0, tol=1e-4, max_iter=100,
coef=None, stderr=None, dispersion=None)
self.model.fit(self.X, self.Y)
# coefficient
print(self.model.coef)
print(self.model.dispersion)
print(self.model.stderr)
print(self.Y - self.model.predict(self.X))
print(self.model.loglike(self.X, self.Y))
self.assertEqual(self.model.coef.shape, (2, 2))
np.testing.assert_array_almost_equal(
self.model.coef,
np.array([[-0.02924966, -0.03484827],
[-0.00978688, 0.00336316]]).reshape(2, -1),
decimal=3)
# std.err of coefficient (calibrated by df_resid)
self.assertEqual(self.model.stderr.shape, (2, 2))
np.testing.assert_array_almost_equal(
self.model.stderr,
np.array([[0.03083908, 0.03121143],
[0.03002101, 0.03038348]]).reshape(2, -1),
decimal=2)
# scale
self.assertEqual(self.model.dispersion.shape, (2, 2))
np.testing.assert_array_almost_equal(
self.model.dispersion,
np.array([[0.94905363, 0.0164185],
[0.0164185, 0.89937019]]),
decimal=3)
# loglike/_per_sample
self.assertAlmostEqual(
self.model.loglike(self.X, self.Y),
-2758.54387369,
places=3)
# to_json
json_dict = self.model.to_json('./tests/linear_models/OLS/MultivariateOLS/')
self.assertEqual(json_dict['properties']['solver'], 'pinv')
# from_json
self.model_from_json = OLS.from_json(json_dict)
np.testing.assert_array_almost_equal(
self.model.coef,
self.model_from_json.coef,
decimal=3)
np.testing.assert_array_almost_equal(
self.model.stderr,
self.model_from_json.stderr,
decimal=3)
np.testing.assert_array_almost_equal(
self.model.dispersion,
self.model_from_json.dispersion,
decimal=3)