def test_to_model_params_conv_net_values_are_close_to_original(self):
model = conv_model(mnist_single_items())
sut, model_weights, is_bias, biases = self.get_sut_and_test_input(
model)
input = self.weights_as_single_vector(model_weights)
recreated_params = sut.to_model_params(input)
recreated_params = self.unwrap_list(recreated_params)
for weights_matrix, is_b, bias, recreated_weights in zip(
model_weights, is_bias, biases, recreated_params):
self.is_close(recreated_weights, is_b, bias, weights_matrix)
def test_two_plots():
model = conv_model(mnist_single_items())
_, _, x_test, y_test = mnist_single_items()
plot_loss_3D(model, ("levels", "3d"), x_test, y_test)
def test_unlearned_conv_model():
model = conv_model(mnist_single_items())
_, _, x_test, y_test = mnist_single_items()
plot_loss_3D(model, "levels", x_test, y_test)
def test_unlearned_conv_model():
model = conv_model(mnist_single_items())
_, _, x_test, y_test = mnist_single_items()
plot_loss(model, (x_test, y_test))
def setup(cls):
if not hasattr(
cls, 'initialized'): # avoid learning networks multiple times
cls.initialized = True
cls.dense = dense_model(toy_dataset())
cls.conv = conv_model(mnist_dataset())
def test_conv_mnist_model():
model = conv_model(mnist_dataset())
_, _, x_test, y_test = mnist_dataset()
plot_loss_3D(model, "levels", x_test, y_test)