parser.add_argument('-m', '--model', help='LSTM type', required=True, type=str, **share_param)
parser.add_argument('-e', '--epoch', help='LSTM type', required=True, type=int, **share_param)
parser.add_argument('-v', '--version', help='', default=None, type=int, **share_param)
return parser.parse_args()
if __name__ == '__main__':
# Ignore warning message by tensor flow
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
# checkpoint
_parser = argparse.ArgumentParser(description='This script is ...', formatter_class=argparse.RawTextHelpFormatter)
args = get_options(_parser)
if args.version is not None:
_checkpoint_dir, _parameter = \
checkpoint_version('./checkpoint/%s' % args.model, version=args.version)
else:
_parameter = toml.load(open('./hyperparameters/%s.toml' % args.model))
_checkpoint_dir, _ = checkpoint_version('./checkpoint/%s' % args.model, _parameter)
# data
raw_train, raw_validation, raw_test, vocab = ptb_raw_data("./simple-examples/data")
iterators = dict()
for raw_data, key in zip([raw_train, raw_validation, raw_test], ["batcher_train", "batcher_valid", "batcher_test"]):
iterators[key] = BatchFeeder(batch_size=_parameter['batch_size'],
num_steps=_parameter['config']['num_steps'],
sequence=raw_data)
model = LanguageModel(max_max_epoch=args.epoch, checkpoint_dir=_checkpoint_dir, **_parameter)
model.train(verbose=True, **iterators)
