def test_gpr():
gpr = IncrementalGaussianProcess()
gpr.first_fit([get_add_skip_model().extract_descriptor()], [0.5])
assert gpr.first_fitted
gpr.incremental_fit([get_concat_skip_model().extract_descriptor()], [0.6])
assert abs(gpr.predict(np.array([get_add_skip_model().extract_descriptor()]))[0] - 0.5) < 1e-4
assert abs(gpr.predict(np.array([get_concat_skip_model().extract_descriptor()]))[0] - 0.6) < 1e-4
def test_gpr():
gpr = IncrementalGaussianProcess()
gpr.first_fit([get_add_skip_model().extract_descriptor()], [0.5])
assert gpr.first_fitted
gpr.incremental_fit([get_concat_skip_model().extract_descriptor()], [0.6])
assert abs(
gpr.predict(np.array([get_add_skip_model().extract_descriptor()]))[0] -
0.5) < 1e-4
assert abs(
gpr.predict(np.array([get_concat_skip_model().extract_descriptor()]))
[0] - 0.6) < 1e-4
def test_dense_wider_stub():
graph = get_add_skip_model()
graph.weighted = False
layer_num = graph.n_layers
graph.to_wider_model(32, 3)
assert graph.n_layers == layer_num
def test_extract_descriptor_add():
descriptor = get_add_skip_model().extract_descriptor()
assert len(descriptor.layers) == 24
assert descriptor.skip_connections == [
(6, 10, NetworkDescriptor.ADD_CONNECT),
(10, 14, NetworkDescriptor.ADD_CONNECT)
]
def test_dense_wider_stub():
model = to_stub_model(get_add_skip_model())
graph = Graph(model)
layer_num = graph.n_layers
graph.to_wider_model(graph.layer_to_id[model.layers[-2]], 3)
assert graph.n_layers == layer_num
def test_dense_wider_stub():
model = get_add_skip_model()
graph = Graph(model, False)
layer_num = graph.n_layers
graph.to_wider_model(19, 3)
assert graph.n_layers == layer_num
def test_dense_wider_stub():
graph = get_add_skip_model()
graph.weighted = False
layer_num = graph.n_layers
graph.to_wider_model(26, 3)
assert graph.n_layers == layer_num
def test_produce_keras_model():
for graph in [get_conv_dense_model(),
get_add_skip_model(),
get_pooling_model(),
get_concat_skip_model()]:
model = graph.produce_keras_model()
assert isinstance(model, keras.models.Model)
def test_keras_model():
for graph in [get_conv_dense_model(),
get_add_skip_model(),
get_pooling_model(),
get_concat_skip_model()]:
keras_model = KerasModel(graph)
keras_model.set_weight_to_graph()
assert isinstance(keras_model, KerasModel)
def test_extract_descriptor_add():
descriptor = get_add_skip_model().extract_descriptor()
assert descriptor.n_conv == 5
assert descriptor.n_dense == 2
assert descriptor.skip_connections == [
(2, 3, NetworkDescriptor.ADD_CONNECT),
(3, 4, NetworkDescriptor.ADD_CONNECT)
]
def test_copy_model():
model = get_add_skip_model()
new_model = Graph(model).produce_model()
input_data = get_conv_data()
output1 = model.predict_on_batch(input_data).flatten()
output2 = new_model.predict_on_batch(input_data).flatten()
assert np.sum(output1 - output2) == 0
def test_gpr():
gpr = IncrementalGaussianProcess()
gpr.first_fit(Graph(get_add_skip_model()).extract_descriptor(), 0.5)
gpr.incremental_fit(
Graph(get_concat_skip_model()).extract_descriptor(), 0.6)
assert abs(
gpr.predict(
np.array([Graph(get_concat_skip_model()).extract_descriptor()]))[0]
- 0.6) < 1e-4
def test_produce_keras_model():
for graph in [
get_conv_dense_model(),
get_add_skip_model(),
get_pooling_model(),
get_concat_skip_model()
]:
model = graph.produce_keras_model()
assert isinstance(model, keras.models.Model)
def test_skip_concatenate():
model = get_add_skip_model()
graph = Graph(model)
new_model = graph.to_concat_skip_model(model.layers[4], model.layers[4])
input_data = get_conv_data()
output1 = model.predict_on_batch(input_data).flatten()
output2 = new_model.predict_on_batch(input_data).flatten()
assert abs(np.sum(output1 - output2)) < 1e-4
def test_dense_wider():
model = get_add_skip_model()
graph = Graph(model)
new_model = graph.to_wider_model(model.layers[-2], 3)
input_data = get_conv_data()
output1 = model.predict_on_batch(input_data).flatten()
output2 = new_model.predict_on_batch(input_data).flatten()
assert abs(np.sum(output1 - output2)) < 1e-4
def test_keras_model():
for graph in [
get_conv_dense_model(),
get_add_skip_model(),
get_pooling_model(),
get_concat_skip_model()
]:
keras_model = KerasModel(graph)
keras_model.set_weight_to_graph()
assert isinstance(keras_model, KerasModel)
def test_dense_wider():
model = get_add_skip_model()
graph = Graph(model, True)
graph.to_wider_model(19, 3)
new_model = graph.produce_model()
input_data = get_conv_data()
output1 = model.predict_on_batch(input_data).flatten()
output2 = new_model.predict_on_batch(input_data).flatten()
assert np.sum(np.abs(output1 - output2)) < 1e-4
def test_dense_wider():
model = get_add_skip_model()
graph = NetworkMorphismGraph(model)
graph.to_wider_model(graph.layer_to_id[model.layers[-2]], 3)
new_model = graph.produce_model()
input_data = get_conv_data()
output1 = model.predict_on_batch(input_data).flatten()
output2 = new_model.predict_on_batch(input_data).flatten()
assert np.sum(np.abs(output1 - output2)) < 1e-4
def test_dense_wider():
graph = get_add_skip_model()
model = graph.produce_model()
graph = deepcopy(graph)
graph.to_wider_model(32, 3)
new_model = graph.produce_model()
input_data = torch.Tensor(get_conv_data())
model.eval()
new_model.eval()
output1 = model(input_data)
output2 = new_model(input_data)
assert (output1 - output2).abs().sum() < 1e-4
def test_dense_wider():
graph = get_add_skip_model()
model = graph.produce_model()
graph = deepcopy(graph)
graph.to_wider_model(26, 3)
new_model = graph.produce_model()
input_data = torch.Tensor(get_conv_data())
model.eval()
new_model.eval()
output1 = model(input_data)
output2 = new_model(input_data)
assert (output1 - output2).abs().sum() < 1e-4
def test_edit_distance():
descriptor1 = get_add_skip_model().extract_descriptor()
descriptor2 = get_concat_skip_model().extract_descriptor()
assert edit_distance(descriptor1, descriptor2) == 12.0
def test_wider_weighted_add():
layer = get_add_skip_model().layers[10]
new_layer = wider_weighted_add(layer, 4)
assert isinstance(new_layer, WeightedAdd)
def test_edit_distance():
descriptor1 = Graph(get_add_skip_model()).extract_descriptor()
descriptor2 = Graph(get_concat_skip_model()).extract_descriptor()
assert edit_distance(descriptor1, descriptor2, 1.0) == 2.0
def test_save_weighted_add():
model = get_add_skip_model()
path = 'tests/resources/temp/m.h5'
model.save(path)
load_model(path, {'WeightedAdd': WeightedAdd})
os.remove(path)
def test_extract_descriptor_add():
descriptor = get_add_skip_model().extract_descriptor()
assert descriptor.n_conv == 5
assert descriptor.n_dense == 2
assert descriptor.skip_connections == [(2, 3, NetworkDescriptor.ADD_CONNECT), (3, 4, NetworkDescriptor.ADD_CONNECT)]
def test_wider_weighted_add():
layer = StubAdd()
layer.set_weights(get_add_skip_model().layers[13].get_weights())
new_layer = wider_weighted_add(layer, 4)
assert isinstance(new_layer, StubAdd)
def test_extract_descriptor_add():
descriptor = get_add_skip_model().extract_descriptor()
assert len(descriptor.layers) == 24
assert descriptor.skip_connections == [(6, 10, NetworkDescriptor.ADD_CONNECT),
(10, 14, NetworkDescriptor.ADD_CONNECT)]
def test_edit_distance():
descriptor1 = get_add_skip_model().extract_descriptor()
descriptor2 = get_concat_skip_model().extract_descriptor()
assert edit_distance(descriptor1, descriptor2) == 12.0
def test_to_stub_model():
model = get_add_skip_model()
stub_model = to_stub_model(model)
assert len(stub_model.layers) == 18
def test_graph_duplicate():
assert same_graph(get_add_skip_model().extract_descriptor(),
get_add_skip_model().extract_descriptor())
assert not same_graph(get_concat_skip_model().extract_descriptor(),
get_add_skip_model().extract_descriptor())
def test_graph_duplicate():
assert same_graph(get_add_skip_model().extract_descriptor(), get_add_skip_model().extract_descriptor())
assert not same_graph(get_concat_skip_model().extract_descriptor(), get_add_skip_model().extract_descriptor())
