def test_fit_on_two_layer_perceptron_is_correct_when_used_with_a_loss(self):
# Given
model = create_2l_perceptron()
maml = MAML(model, loss=tf.keras.losses.MSE)
data_x = np.ones((1,))
data_y = np.zeros((1, 1))
expected_weights = [
-3 * np.ones((1, 2)),
-3 * np.ones((2, 1)),
]
# When
eval_model = maml.fit(data_x, data_y)
weight_set = [layer.kernel for layer in eval_model.layers]
# Then
for i_weight, weights in enumerate(weight_set):
self.assertTrue(np.all(weights == expected_weights[i_weight]))
def test_fit_on_two_layer_perceptron_is_correct_when_passed_a_value_for_alpha(self):
# Given
model = create_2l_perceptron()
maml = MAML(model, loss=lambda y, p: p)
data_x = np.ones((1,))
data_y = np.zeros((1, 1))
expected_weights = [
0.5 * np.ones((1, 2)),
0.5 * np.ones((2, 1)),
]
# When
eval_model = maml.fit(data_x, data_y, alpha=0.5)
weight_set = [layer.kernel for layer in eval_model.layers]
# Then
for i_weight, weights in enumerate(weight_set):
self.assertTrue(np.all(weights == expected_weights[i_weight]))