def assert_models_same(self, model, model2):
"""Assert two keras models same."""
if model.optimizer:
assert (
model.optimizer.get_config() == model2.optimizer.get_config())
assert len(model.layers) == len(model2.layers) # shallow comparison
layers = list(iterlayers(model))
layers2 = list(iterlayers(model2))
assert len(layers) == len(layers2) # deep comparison
def test__freeze_sets_trainable_except(self, multinetwork):
model = 'forecaster'
all_models = multinetwork.model_names
multinetwork._freeze_models_except(model)
for m in all_models:
if m == model:
for layer in iterlayers(multinetwork.model[m]):
assert layer.trainable is True
else:
for layer in iterlayers(multinetwork.model[m]):
assert layer.trainable is False
def test_multimodel(self, multimodel, multimodel_num_layers):
for name, model in multimodel.items():
layers = list(iterlayers(model))
expected_num_layers = multimodel_num_layers[name]
assert len(layers) == expected_num_layers
def test_model(self, model, n_layers):
layers = list(iterlayers(model))
assert len(layers) == n_layers
def test__freeze_sets_trainable_all(self, multinetwork):
all_models = multinetwork.model_names
multinetwork._freeze_models_except(all_models)
for m in all_models:
for layer in iterlayers(multinetwork.model[m]):
assert layer.trainable is True
def test__freeze_sets_trainable_none(self, multinetwork, model):
multinetwork._freeze_models_except(model)
all_models = multinetwork.model_names
for m in all_models:
for layer in iterlayers(multinetwork.model[m]):
assert layer.trainable is False