def main():
parser = opts.get_parser()
args = parser.parse_args()
params = read_params(args)
out_dir = make_and_fill_output_dir(args, params)
load_unspecified_params(params)
# Initialize model and move it to gpu
torch.cuda.set_device(args.gpu)
model = gen_model(params, args)
print(model)
model.cuda(args.gpu)
# Load the datasets and setup optimizer
datasets, dataset_stats = load_data(args.data)
optimizer = setup_optimizer(model, params)
if args.eval:
dev_epoch(model=model,
data_iter=datasets["dev"].get_batched_iter(
params["batch_size"]),
desc="Evaluation:",
data_iter_len=dataset_stats["dev"]["words"] //
params["batch_size"])
else:
n_epochs = int(args.n_epochs)
for i in range(1, n_epochs + 1):
run_epoch(model=model,
optimizer=optimizer,
datasets=datasets,
dataset_stats=dataset_stats,
out_dir=out_dir,
batch_size=params["batch_size"],
bptt_len=params["bptt_len"],
epoch_num=i)
def main():
parser = opts.get_parser()
args = parser.parse_args()
params = read_params(args)
make_and_fill_output_dir(args, params)
load_unspecified_params(params)
# Initialize model and move it to gpu
torch.cuda.set_device(args.gpu)
model = gen_model(params, args.resume_from)
model.cuda(args.gpu)
model.init_params()
# Load the datasets and setup optimizer
datasets, dataset_stats = load_data(args.data)
optimizer = setup_optimizer(model)
n_epochs = int(args.n_epochs)
for i in range(1, n_epochs + 1):
run_epoch(model=model,
optimizer=optimizer,
datasets=datasets,
dataset_stats=dataset_stats,
batch_size=params["batch_size"],
bptt_len=params["bptt_len"],
epoch_num=i)
def main():
parser = opts.get_parser()
args = parser.parse_args()
params = read_params(args)
make_and_fill_output_dir(args, params)
load_unspecified_params(params)
# Initialize model and move it to gpu
model = gen_model(params, args.resume_from)
model_to_gpu(model, args.gpu)
# Load the datasets and setup optimizer
datasets, dataset_stats = load_data(args.data)
optimizer = setup_optimizer(model)
n_epochs = int(args.n_epochs)
for i in range(1, n_epochs + 1):
run_epoch(model=model,
optimizer=optimizer,
datasets=datasets,
dataset_stats=dataset_stats,
batch_size=params["batch_size"],
bptt_len=params["bptt_len"],
epoch_num=i)
save_model(path.join(args.out, "snapshot_{0}".format(i)), model)
def main():
parser = get_parser()
p = parser.parse_args()
if p.config is not None:
with open(p.config, 'r') as f:
default_arg = yaml.load(f)
key = vars(p).keys()
for k in default_arg.keys():
if k not in key:
print('Wrong Arg: {}'.format(k))
assert (k in key)
parser.set_defaults(**default_arg)
args = parser.parse_args()
processor = Processor(args)
processor.start()
def main(opts):
"""docstring for main"""
if opts.mode == "colab_tpu":
strategy = start_tpu()
t = trainer(opts, strategy)
t.train()
elif opts.mode == "gpu":
opts.shuffle_buffer = 64
opts.val_batch_size = 32
opts.train_batch_size = 32
strategy = tf.distribute.get_strategy()
t = trainer(opts, strategy)
t.train()
elif opts.mode == "debug":
opts.shuffle_buffer = None
opts.val_batch_size = 1
opts.train_batch_size = 1
strategy = tf.distribute.get_strategy()
t = trainer(opts, strategy)
t.train()
if __name__ == "__main__":
opts = get_parser()
num_data = opts.tfrecord_path.split('/')[-1].split('.')[0].split('_')[-1]
opts.num_data = int(num_data)
opts.weight_path = opts.weight_path + opts.exp_name
main(opts)