dim_embedding=args.embedding_dim,
vocab_size=train_data_engine.tokenizer.get_vocab_size(),
attr_vocab_size=attr_vocab_size,
n_layers=args.n_layers,
bidirectional=args.bidirectional,
model_dir=args.model_dir,
log_dir=args.log_dir,
is_load=args.is_load,
replace_model=args.replace_model,
device=device,
dir_name=args.dir_name,
with_intent=args.with_intent)
lm = None
if args.lm_model_dir:
lm = LM.load_pretrained(args.lm_model_dir, train_data_engine,
test_data_engine, device)
made = None
if args.made_model_dir:
# made = Marginal.load_pretrained(
# args.made_model_dir,
# device=device
# )
made = Marginal(
batch_size=args.batch_size,
optimizer=args.optimizer,
learning_rate=args.learning_rate,
train_data_engine=train_data_engine,
test_data_engine=test_data_engine,
dim_hidden=args.hidden_size,
dim_embedding=args.embedding_dim,
regen=args.regen,
train=False)
vocab, rev_vocab, token_vocab, rev_token_vocab = \
pickle.load(open(args.vocab_file, 'rb'))
attr_vocab_size = len(token_vocab)
vocab_size = args.vocab_size + 4
model = LM(batch_size=args.batch_size,
optimizer=args.optimizer,
learning_rate=args.learning_rate,
train_data_engine=train_data_engine,
test_data_engine=test_data_engine,
dim_hidden=args.hidden_size,
dim_embedding=args.embedding_dim,
vocab_size=vocab_size,
n_layers=args.n_layers,
model_dir=args.model_dir,
log_dir=args.log_dir,
is_load=args.is_load,
replace_model=args.replace_model,
device=device,
dir_name=args.dir_name)
# record model config
if not args.is_load:
with open(os.path.join(model.model_dir, "lm_config"), "w+") as f:
for arg in vars(args):
f.write("{}: {}\n".format(arg, str(getattr(args, arg))))
f.close()