config['train']['max_iter'] = args.max_iter
if args.batch_size > 0:
config['train']['batch_size'] = args.batch_size
num_classes = int(config['model']['num_classes'])
batch_size = int(config['train']['batch_size'])
train_data = VFDataset([1, 3, 5, 8])
valid_data = VFDataset([4, 6])
test_data = VFDataset([2, 7])
class_weights = get_class_weights(num_classes=num_classes,
train_data=train_data,
valid_data=valid_data)
train_loader = DataLoader(train_data, batch_size=batch_size, shuffle=True)
valid_loader = DataLoader(valid_data, batch_size=batch_size, shuffle=False)
test_loader = DataLoader(test_data, batch_size=batch_size, shuffle=False)
device = torch.device('cuda')
Model = model_dict[args.model]
model = Model(n_classes=num_classes).to(device)
model = nn.DataParallel(model)
trainer = Trainer(model, config, train_loader, valid_loader, class_weights,
device)
trainer.train(vis_data=test_data)
accuracy, precisions, recalls, IoUs = trainer.eval(test_loader)
print('accuracy: {:.3f}'.format(accuracy))
print_stats(precisions, recalls, IoUs)
max_candidates = 80
max_candidates_len = 10
logger = preprocess.config_logger('main')
model_path = 'models/'
if not os.path.exists(model_path):
os.makedirs(model_path)
tokenize = TweetTokenizer().tokenize
# if not os.path.isfile('{}.preprocessed.pickle'.format(in_file)) \
# or not os.path.isfile('{}.preprocessed.pickle'.format(dev_file)):
# logger.error("train file or dev file is not exists.")
# exit()
if not evaluation_mode:
logger.info('Load preprocessed train and dev file.')
dataset = MyDataset(in_file, range(10), use_elmo, max_nodes,
max_query_size, max_candidates, max_candidates_len)
dev_dataset = MyDataset(dev_file, range(25, 30), use_elmo, max_nodes,
max_query_size, max_candidates,
max_candidates_len)
logger.info("Data has prepared, train: %d, dev: %d." %
(len(dataset), len(dev_dataset)))
trainer = Trainer(dataset, dev_dataset, logger)
trainer.train()
else:
logger.info('Load preprocessed evaluation data file.')
dataset = MyDataset(dev_file, use_elmo, max_nodes, max_query_size,
max_candidates, max_candidates_len)
trainer = Trainer(dataset, dataset)
trainer.eval()
return test_dataset
if __name__ == "__main__":
args = load_from_pickle('./checkpoint/config.pt')
logger = get_logger(
args.log_dir +
"NER_Test_{}.txt".format(time.strftime("%m-%d_%H-%M-%S")))
logger.info(args)
logger.info('load embedding......')
embedding = c_data_utils.load_embedding(args, args.vocab)
embedding = torch.tensor(embedding, dtype=torch.float)
logger.info('load embedding done')
trainer = Trainer(args, w_embedding=embedding)
trainer.load()
test_dataset = load_data(args, args.vocab)
test_loader = DataLoader(dataset=test_dataset,
batch_size=args.batch_size,
shuffle=False)
weighted_f1, macro_f1, p, r, acc, during_time, pred_labels, gold_labels = trainer.eval(
test_dataset)
logger.info(
"Test Acc: {:.4f} P: {:.4f} R: {:.4f} F1:{:.4f} Time: {}".format(
acc, p, r, weighted_f1, int(during_time)))
logger.info(classification_report(gold_labels, pred_labels))
