def test_fit(deepspeech: DeepSpeech, generator: DataGenerator,
config_path: str, alphabet_path: str, model_dir: str):
# Test save best weights (overwrite the best result)
weights_path = os.path.join(model_dir, 'weights_copy.hdf5')
deepspeech.save(weights_path)
distributed_weights = deepspeech.distributed_model.get_weights()
model_checkpoint = deepspeech.callbacks[2]
model_checkpoint.best_result = 0
model_checkpoint.best_weights_path = weights_path
history = deepspeech.fit(train_generator=generator,
dev_generator=generator,
epochs=2,
shuffle=False)
assert type(history) == History
# Test the returned model has `test_weights`
deepspeech_weights = deepspeech.model.get_weights()
new_deepspeech = DeepSpeech.construct(config_path, alphabet_path)
new_deepspeech.load(model_checkpoint.best_weights_path)
new_deepspeech_weights = new_deepspeech.model.get_weights()
assert is_same(deepspeech_weights, new_deepspeech_weights)
# Test that distributed model appropriate update weights
new_distributed_weights = deepspeech.distributed_model.get_weights()
assert is_same(distributed_weights, new_distributed_weights)
def test_save_load(deepspeech: DeepSpeech, config: Configuration,
config_path: str, alphabet_path: str, model_dir: str):
weights_path = os.path.join(model_dir, 'weights.hdf5')
model_weights = deepspeech.model.get_weights()
deepspeech.save(weights_path)
new_deepspeech = DeepSpeech.construct(config_path, alphabet_path)
new_deepspeech.model = deepspeech.get_model(**config.model,
is_gpu=False,
random_state=123)
new_model_weights = new_deepspeech.model.get_weights()
assert not is_same(model_weights, new_model_weights)
new_deepspeech.load(weights_path)
new_model_weights = new_deepspeech.model.get_weights()
assert is_same(model_weights, new_model_weights)
def load(name: str):
from deepspeech import DeepSpeech
if os.path.isdir(name):
model_dir = name
else:
model_dir = get_pretrained_model_dir(name)
config_path = os.path.join(model_dir, 'configuration.yaml')
alphabet_path = os.path.join(model_dir, 'alphabet.txt')
weights_path = os.path.join(model_dir, 'weights.hdf5')
assert os.path.isfile(config_path), 'The config file required in the directory'
assert os.path.isfile(alphabet_path), 'The alphabet file required in the directory'
deepspeech = DeepSpeech.construct(config_path, alphabet_path)
if os.path.isfile(weights_path):
deepspeech.load(weights_path)
return deepspeech
def deepspeech(config_path: str, alphabet_path: str) -> DeepSpeech:
return DeepSpeech.construct(config_path, alphabet_path)