def main(_) -> None:
with testing_utils.model_type_scope('functional'):
model = testing_utils.get_small_mlp(1, 4, input_dim=3)
model.layers[-1].activity_regularizer = regularizers.get('l2')
model.activity_regularizer = regularizers.get('l2')
model.compile(loss='mse', optimizer='rmsprop')
def callable_loss():
return tf.reduce_sum(model.weights[0])
model.add_loss(callable_loss)
print(f'_____Writing saved model to: {FLAGS.output_path}')
model.save(FLAGS.output_path)
def _test_sequential_model_type(f, test_or_class, *args, **kwargs):
with testing_utils.model_type_scope('sequential'):
f(test_or_class, *args, **kwargs)
def _test_functional_model_type(f, test_or_class, *args, **kwargs):
with testing_utils.model_type_scope('functional'):
f(test_or_class, *args, **kwargs)
def _test_subclass_model_type(f, test_or_class, *args, **kwargs):
with testing_utils.model_type_scope('subclass'):
f(test_or_class, *args, **kwargs)
def context_managers(self, kwargs):
model_type = kwargs.pop('model_type', None)
if model_type in KERAS_MODEL_TYPES:
return [testing_utils.model_type_scope(model_type)]
else:
return []
