def main():
parser = ArgumentParser()
parser.add_argument('mode',
choices=['train', 'predict'],
help='pipeline mode')
parser.add_argument('model',
choices=['rnn', 'cnn', 'multihead'],
help='model to be used')
parser.add_argument('dataset',
choices=['QQP', 'SNLI'],
help='dataset to be used')
parser.add_argument('--gpu',
default='0',
help='index of GPU to be used (default: %(default))')
args = parser.parse_args()
set_visible_gpu(args.gpu)
main_config = init_config()
model_config = init_config(args.model)
mode = args.mode
if 'train' in mode:
train(main_config, model_config, args.model, args.dataset)
else:
predict(main_config, model_config, args.model)
def main():
parser = ArgumentParser()
parser.add_argument('--data-dir',
default='./corpora',
help='Path to original quora split')
parser.add_argument('--model-dir',
default='./model_dir',
help='Path to save the trained model')
parser.add_argument('--use-help',
choices=[True, False],
default=False,
type=bool,
help='should model use help on difficult examples')
parser.add_argument('--gpu',
default='0',
help='index of GPU to be used (default: %(default))')
parser.add_argument('--embeddings',
choices=['no', 'fixed', 'tunable'],
default='no',
type=str,
help='embeddings')
parser.add_argument('--batch-size',
choices=[4, 128, 256, 512],
default=128,
type=int,
help='batch size')
parser.add_argument('--syn-weight',
default=1,
type=float,
help='Weight for loss function')
args = parser.parse_args()
logger.info(args)
if args.embeddings == 'no':
args.use_embed = False
args.tune = False
else:
args.use_embed = True
args.tune = False if args.embeddings == 'fixed' else True
set_visible_gpu(args.gpu)
args.model_dir = '{}_bilstm_{}_{}_{}'.format(args.model_dir,
args.embeddings,
args.batch_size,
args.syn_weight
)
main_config = init_config()
args.max_seq_length, args.vocab_size = get_vocab(main_config, args, logger)
logger.info(args)
train(main_config, args)
def main():
parser = ArgumentParser()
parser.add_argument(
'mode',
choices=['train', 'predict'],
help='pipeline mode',
)
parser.add_argument(
'model',
choices=['rnn', 'cnn', 'multihead'],
help='model to be used',
)
parser.add_argument(
'dataset',
choices=['QQP', 'SNLI', 'ANLI'],
nargs='?',
help='dataset to be used',
)
parser.add_argument(
'--experiment_name',
required=False,
help='the name of run experiment',
)
parser.add_argument(
'--gpu',
default='0',
help='index of GPU to be used (default: %(default))',
)
args = parser.parse_args()
if 'train' in args.mode:
if args.dataset is None:
parser.error('Positional argument [dataset] is mandatory')
set_visible_gpu(args.gpu)
main_config = init_config()
model_config = init_config(args.model)
mode = args.mode
experiment_name = args.experiment_name
if experiment_name is None:
experiment_name = create_experiment_name(args.model, main_config,
model_config)
if 'train' in mode:
train(main_config, model_config, args.model, experiment_name,
args.dataset)
else:
predict(main_config, model_config, args.model, experiment_name)