def test_optimizer(optimizer1):
l0 = InputLayer(2)
l1 = DenseLayer(2, 3, 'relu')
l1.W = W1
l1.b = b1
layers = [l0, l1]
gradients = [{'W': dJdW1, 'b': dJdb1}]
expected_optimized_W1 = optimized_W1
expected_optimized_b1 = optimized_b1
optimizer1.reset(layers)
optimizer1.optimize(layers, gradients)
assert_array_almost_equal(l1.params['W'], expected_optimized_W1, decimal=4)
assert_array_almost_equal(l1.params['b'], expected_optimized_b1, decimal=4)
def add_dense_layer(self, dimension: int, activation_function_name: str):
"""
Add a dense layer to this DockNet after the last layer
:param dimension: number of neurons of the layer to add
:param activation_function_name: name of the activation function to use in this layer
"""
layer = DenseLayer(self.layers[-1].dimension, dimension,
activation_function_name)
self.layers.append(layer)
def test_initialize(initializer1):
np.random.seed(1)
expected_W1 = [[0.08121727, -0.03058782], [-0.02640859, -0.05364843], [0.04327038, -0.11507693]]
expected_b1 = [[0.08724059], [-0.03806035], [0.01595195]]
l0 = InputLayer(2)
l1 = DenseLayer(2, 3, 'relu')
layers = [l0, l1]
initializer1.initialize(layers)
assert l0.params == {}
assert_array_almost_equal(l1.params['W'], expected_W1)
assert_array_almost_equal(l1.params['b'], expected_b1)
def layer1() -> DenseLayer:
l1 = DenseLayer(2, 3, 'relu')
l1.W = W1
l1.b = b1
yield l1