from utils import torch_utils, helper
parser = argparse.ArgumentParser()
parser.add_argument('--data_dir', type=str, default='dataset/Laptops')
parser.add_argument('--vocab_dir', type=str, default='dataset/Laptops')
parser.add_argument('--batch_size', type=int, default=32, help='Batch size.')
parser.add_argument('--lower', default=True, help='Lowercase all words.')
parser.add_argument('--model_dir',
type=str,
default='saved_models/best_model.pt',
help='Directory of the model.')
args = parser.parse_args()
print("Loading model from {}".format(args.model_dir))
opt = torch_utils.load_config(args.model_dir)
loaded_model = GCNTrainer(opt)
loaded_model.load(args.model_dir)
print("Loading vocab...")
token_vocab = Vocab.load_vocab(args.vocab_dir + '/vocab_tok.vocab') # token
post_vocab = Vocab.load_vocab(args.vocab_dir + '/vocab_post.vocab') # position
pos_vocab = Vocab.load_vocab(args.vocab_dir + '/vocab_pos.vocab') # POS
dep_vocab = Vocab.load_vocab(args.vocab_dir + '/vocab_dep.vocab') # deprel
pol_vocab = Vocab.load_vocab(args.vocab_dir + '/vocab_pol.vocab') # polarity
vocab = (token_vocab, post_vocab, pos_vocab, dep_vocab, pol_vocab)
print(
"token_vocab: {}, post_vocab: {}, pos_vocab: {}, dep_vocab: {}, pol_vocab: {}"
.format(len(token_vocab), len(post_vocab), len(pos_vocab), len(dep_vocab),
len(pol_vocab)))
print("Loading data from {} with batch size {}...".format(
print("Loading data from {} with batch size {}...".format(
args.dataset, args.batch_size))
train_batch = DataLoader('./dataset/' + args.dataset + '/train.json',
args.batch_size, args, dicts)
test_batch = DataLoader('./dataset/' + args.dataset + '/test.json',
args.batch_size, args, dicts)
# create the folder for saving the best models and log file
model_save_dir = args.save_dir
helper.ensure_dir(model_save_dir, verbose=True)
file_logger = helper.FileLogger(
model_save_dir + '/' + args.log,
header="# epoch\ttrain_loss\ttest_loss\ttrain_acc\ttest_acc\ttest_f1")
# create the model
trainer = GCNTrainer(args, emb_matrix=emb_matrix)
# start training
train_acc_history, train_loss_history, test_loss_history, f1_score_history = [], [], [], [0.]
test_acc_history = [0.]
for epoch in range(1, args.num_epoch + 1):
train_loss, train_acc, train_step = 0., 0., 0
for i, batch in enumerate(train_batch):
loss, acc = trainer.update(batch)
train_loss += loss
train_acc += acc
train_step += 1
if train_step % args.log_step == 0:
print("train_loss: {}, train_acc: {}".format(
train_loss / train_step, train_acc / train_step))
args.data_dir, args.batch_size))
train_batch = DataLoader(args.data_dir + '/train.json', args.batch_size, args,
vocab)
test_batch = DataLoader(args.data_dir + '/test.json', args.batch_size, args,
vocab)
# check saved_models director
model_save_dir = args.save_dir
helper.ensure_dir(model_save_dir, verbose=True)
# log
file_logger = helper.FileLogger(
model_save_dir + '/' + args.log,
header="# epoch\ttrain_loss\ttest_loss\ttrain_acc\ttest_acc\ttest_f1")
# build model
trainer = GCNTrainer(args, emb_matrix=word_emb)
# start training
train_acc_history, train_loss_history, test_loss_history, f1_score_history = [], [], [], [0.]
test_acc_history = [0.]
for epoch in range(1, args.num_epoch + 1):
train_loss, train_acc, train_step = 0., 0., 0
for i, batch in enumerate(train_batch):
loss, acc = trainer.update(batch)
train_loss += loss
train_acc += acc
train_step += 1
if train_step % args.log_step == 0:
print("train_loss: {}, train_acc: {}".format(
train_loss / train_step, train_acc / train_step))