def main():
# save file
config.mulu = datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
config.save_dir = os.path.join(config.save_direction, config.mulu)
if not os.path.isdir(config.save_dir):
os.makedirs(config.save_dir)
# get iter
create_alphabet = None
train_iter, dev_iter, test_iter, create_alphabet = load_Data(config)
config.embed_num = create_alphabet.word_alphabet.vocab_size
config.class_num = create_alphabet.label_alphabet.vocab_size
config.paddingId = create_alphabet.word_paddingId
config.label_paddingId = create_alphabet.label_paddingId
config.create_alphabet = create_alphabet
print("embed_num : {}, class_num : {}".format(config.embed_num, config.class_num))
print("PaddingID {}".format(config.paddingId))
if config.pretrained_embed:
print("Using Pre_Trained Embedding.")
pretrain_embed = load_pretrained_emb_zeros(path=config.pretrained_embed_file,
text_field_words_dict=create_alphabet.word_alphabet.id2words,
pad=paddingkey)
config.pretrained_weight = pretrain_embed
model = None
if config.model_BiLstm is True:
print("loading model.....")
model = BiLSTM(config)
print(model)
if config.use_cuda is True:
model = model.cuda()
print("Training Start......")
train.train(train_iter=train_iter, dev_iter=dev_iter, test_iter=test_iter, model=model, config=config)
def main():
# save file
mulu = datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
args.mulu = mulu
args.save_dir = os.path.join(args.save_dir, mulu)
if not os.path.isdir(args.save_dir):
os.makedirs(args.save_dir)
# get iter
train_iter, dev_iter, test_iter, create_alphabet = load_Conll2003_NER(args)
args.embed_num = create_alphabet.word_alphabet.vocab_size
args.class_num = create_alphabet.label_alphabet.vocab_size
args.paddingId = create_alphabet.word_paddingId
args.create_alphabet = create_alphabet
print("embed_num : {}, class_num : {}".format(args.embed_num,
args.class_num))
print("PaddingID {}".format(args.paddingId))
if args.word_Embedding:
print("Using Pre_Trained Embedding.")
pretrain_embed = load_pretrained_emb_avg(
path=args.word_Embedding_Path,
text_field_words_dict=create_alphabet.word_alphabet.id2words,
pad=paddingkey)
args.pretrained_weight = pretrain_embed
# print params
show_params()
model = None
if args.BiLSTM is True:
print("loading BiLSTM model.....")
model = BiLSTM(args)
shutil.copy("./models/BiLSTM.py", args.save_dir)
print(model)
elif args.BiLSTM_CRF is True:
print("loading BiLSTM_CRF model......")
bilstm = BiLSTM(args)
crf = CRF(args)
model = BiLSTM_CRF(BiLSTM=bilstm, CRF=crf, args=args)
print(model)
if args.use_cuda is True:
print("Using Cuda To Speed Up......")
model = model.cuda()
if os.path.exists("./Test_Result.txt"):
os.remove("./Test_Result.txt")
print("Training Start......")
if args.BiLSTM is True:
train_conll2003.train(train_iter=train_iter,
dev_iter=dev_iter,
test_iter=test_iter,
model=model,
args=args)
elif args.BiLSTM_CRF is True:
train_conll2003_CRF.train(train_iter=train_iter,
dev_iter=dev_iter,
test_iter=test_iter,
model=model,
args=args)
dropout = args.dropout
# tag
out_dim = len(tag2id) + 2
model = BiLSTM(train_vocab_size, embedding_dim, num_layers, hidden_dim,
dropout, out_dim)
optimizer = optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
batch_size = args.batch_size
if args.cuda:
model.cuda()
train_word_id_lists = train_word_id_lists.cuda()
train_tag_id_lists = train_tag_id_lists.cuda()
dev_word_id_lists = dev_word_id_lists.cuda()
dev_tag_id_lists = dev_tag_id_lists.cuda()
test_word_id_lists = test_word_id_lists.cuda()
test_tag_id_lists = test_tag_id_lists.cuda()
train_total_step = train_vocab_size // batch_size + 1
dev_total_step = dev_vocab_size // batch_size + 1
test_total_step = dev_vocab_size // batch_size + 1
best_model = None
best_val_loss = float('inf')