def test_default_output(regtest):
"""default output should not change"""
fv3fit.set_random_seed(0)
config = fv3fit.ConvolutionalNetworkConfig()
array = np.random.randn(2, 10, 10, 3)
out = config.build(array, n_features_out=3)
print_result(out, decimals=4, file=regtest)
def test_network_has_correct_number_of_hidden_layers(depth):
fv3fit.set_random_seed(0)
config = fv3fit.ConvolutionalNetworkConfig(depth=depth)
n_features_in, n_features_out = 5, 5
input = tf.keras.layers.Input(shape=(10, 10, n_features_in))
convolutional_network = config.build(input, n_features_out=n_features_out)
# one layer of depth is the output
assert len(convolutional_network.hidden_outputs) == depth - 1
def test_output_is_correct_shape(input_shape, kernel_size, depth, features_out,
base_output_shape):
config = fv3fit.ConvolutionalNetworkConfig(kernel_size=kernel_size,
depth=depth)
array = np.random.randn(*input_shape)
convolutional_network = config.build(array, n_features_out=features_out)
output_shape = tuple(base_output_shape) + (features_out, )
assert convolutional_network.output.shape == output_shape
def test_output_type():
config = fv3fit.ConvolutionalNetworkConfig()
array = np.random.randn(2, 10, 10, 3)
convolutional_network = config.build(array, n_features_out=3)
assert isinstance(convolutional_network, fv3fit.ConvolutionalNetwork)
assert isinstance(convolutional_network.output, tf.Tensor)
assert isinstance(convolutional_network.hidden_outputs, Sequence)
assert all(
isinstance(item, tf.Tensor)
for item in convolutional_network.hidden_outputs)
def test_network_has_gaussian_noise_layer():
fv3fit.set_random_seed(0)
config = fv3fit.ConvolutionalNetworkConfig(gaussian_noise=0.1)
n_features_in, n_features_out = 5, 5
input = tf.keras.layers.Input(shape=(10, 10, n_features_in))
convolutional_network = config.build(input, n_features_out=n_features_out)
model = tf.keras.Model(inputs=input, outputs=convolutional_network.output)
assert any(
isinstance(layer, tf.keras.layers.GaussianNoise)
for layer in model.layers)
def test_hidden_layers_have_samples_equal_to_filters(depth, filters):
input_shape = (3, 10, 10, 5)
kernel_size = 3
features_out = 1
config = fv3fit.ConvolutionalNetworkConfig(kernel_size=kernel_size,
depth=depth,
filters=filters)
array = np.random.randn(*input_shape)
convolutional_network = config.build(array, n_features_out=features_out)
for output in convolutional_network.hidden_outputs:
assert output.shape[-1] == filters
def test_standard_input_gives_standard_output():
fv3fit.set_random_seed(0)
config = fv3fit.ConvolutionalNetworkConfig()
array = np.random.randn(2, 10, 10, 20)
np.testing.assert_almost_equal(np.mean(array), 0.0, decimal=1)
np.testing.assert_almost_equal(np.std(array), 1.0, decimal=1)
convolutional_network = config.build(array, n_features_out=3)
np.testing.assert_almost_equal(np.mean(convolutional_network.output),
0.0,
decimal=1)
out_std = np.std(convolutional_network.output)
# std isn't going to be 1 because of relu activation function, should be less
assert out_std < 1.0
assert out_std > 0.1
def test_modifying_one_input_modifies_correct_number_of_outputs(
kernel_size, depth, n_expected_changes):
"""
Test that when you modify an input in the center of the domain,
the correct number of output values change for the initialized network.
"""
features_out = 1
input_shape = (1, 17, 17, 1)
config = fv3fit.ConvolutionalNetworkConfig(kernel_size=kernel_size,
depth=depth)
array = np.random.randn(*input_shape)
input_layer = tf.keras.layers.Input(shape=input_shape[1:])
convolutional_network = config.build(input_layer,
n_features_out=features_out)
model = tf.keras.Model(inputs=input_layer,
outputs=convolutional_network.output)
first_output = model.predict(array)
array[0, 8, 8, 0] = np.random.randn()
second_output = model.predict(array)
assert np.sum(first_output != second_output) == n_expected_changes