def test_regularized_model_selection_lambda(self):
x_train = TEST_DATA['rms']['x_train']
y_train = TEST_DATA['rms']['y_train']
x_val = TEST_DATA['rms']['x_val']
y_val = TEST_DATA['rms']['y_val']
M = TEST_DATA['rms']['M']
lambda_values = TEST_DATA['rms']['lambda_values']
lambda_org = TEST_DATA['rms']['lambda']
_, _, _, lambda_computed = regularized_model_selection(x_train, y_train, x_val, y_val, M, lambda_values)
self.assertAlmostEqual(lambda_org, lambda_computed, 6)
def test_regularized_model_selection_val_err(self):
x_train = TEST_DATA['rms']['x_train']
y_train = TEST_DATA['rms']['y_train']
x_val = TEST_DATA['rms']['x_val']
y_val = TEST_DATA['rms']['y_val']
M = TEST_DATA['rms']['M']
lambda_values = TEST_DATA['rms']['lambda_values']
val_err = TEST_DATA['rms']['val_err']
_, _, val_err_computed, _ = regularized_model_selection(x_train, y_train, x_val, y_val, M, lambda_values)
self.assertAlmostEqual(val_err, val_err_computed, 6)
def test_regularized_model_selection_w(self):
x_train = TEST_DATA['rms']['x_train']
y_train = TEST_DATA['rms']['y_train']
x_val = TEST_DATA['rms']['x_val']
y_val = TEST_DATA['rms']['y_val']
M = TEST_DATA['rms']['M']
lambda_values = TEST_DATA['rms']['lambda_values']
w = TEST_DATA['rms']['w']
w_computed, _, _, _ = regularized_model_selection(x_train, y_train, x_val, y_val, M, lambda_values)
max_diff = np.max(np.abs(w - w_computed))
self.assertAlmostEqual(max_diff, 0, 6)
def test_regularized_model_selection_w(self):
x_train = TEST_DATA['rms']['x_train']
y_train = TEST_DATA['rms']['y_train']
x_val = TEST_DATA['rms']['x_val']
y_val = TEST_DATA['rms']['y_val']
M = TEST_DATA['rms']['M']
lambda_values = TEST_DATA['rms']['lambda_values']
w_expected = TEST_DATA['rms']['w']
w, _, _, _ = regularized_model_selection(x_train, y_train, x_val, y_val, M, lambda_values)
self.assertEqual(np.shape(w), (8, 1))
np.testing.assert_almost_equal(w, w_expected)
def test_regularized_model_selection_lambda(self):
x_train = TEST_DATA['rms']['x_train']
y_train = TEST_DATA['rms']['y_train']
x_val = TEST_DATA['rms']['x_val']
y_val = TEST_DATA['rms']['y_val']
M = TEST_DATA['rms']['M']
lambda_values = TEST_DATA['rms']['lambda_values']
lambda_expected = TEST_DATA['rms']['lambda']
_, _, _, lambda_ = regularized_model_selection(
x_train, y_train, x_val, y_val, M, lambda_values)
self.assertEqual(np.size(lambda_), 1)
self.assertAlmostEqual(lambda_, lambda_expected)
plt.tight_layout()
plt.draw()
print('\n--- Wcisnij klawisz, aby kontynuowac ---')
plt.waitforbuttonpress(0)
print(
'\n--- Selekcja modelu dla liniowego zadania najmniejszych kwadratow z regularyzacja ---'
)
print(
'-- Stopien M=7. Liczba punktow treningowych N=50. Liczba punktow walidacyjnych N=20 --'
)
M = 7
lambdas = [0.001, 0.003, 0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100, 300]
w, train_err, val_err, best_lambda = regularized_model_selection(
data['x_train_50'], data['y_train_50'], data['x_val_20'],
data['y_val_20'], M, lambdas)
y_model = polynomial(x_plot, w)
fig = plt.figure(figsize=(6, 5),
num='Selekcja modelu dla parametru regularyzacji')
sub = fig.add_subplot(1, 1, 1)
sub.set_title('M={} Najlepsze $\lambda$={}'.format(M, best_lambda))
plot_model(data['x_train_50'], data['y_train_50'], x_plot, y_obj, y_model,
data['x_val_20'], data['y_val_20'], train_err, val_err)
plt.tight_layout()
plt.draw()
print('\n--- Wcisnij klawisz, aby kontynuowac ---')
plt.waitforbuttonpress(0)