if __name__ == '__main__':
parser = argparse.ArgumentParser(description='TRAIN')
parser.add_argument('--num', type=int)
parser.add_argument('--embed', type=int)
parser.add_argument('--units', type=int)
parser.add_argument('--epoch', type=int)
parser.add_argument('--gpu', type=int)
parser.add_argument('--save', type=str)
parser.add_argument('--batch', type=int, default=2)
args = parser.parse_args()
gpu_config(args.gpu)
model, (train_x, train_y), (test_x,
test_y) = bilstm_crf_model.create_model(
args.embed, args.units)
# used for multi checkpoints to vote
#filepath = args.save+'/weights-improvement-{epoch:02d}-{val_acc:.4f}.h5'
# only get the best single model
filepath = args.save + '/model.h5'
checkpoint = ModelCheckpoint(filepath,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
model.fit(train_x,
train_y,
batch_size=args.batch,
len(predict_text))
raw = model.predict(str_)[0][-length:]
result = [np.argmax(row) for row in raw]
result_tags = [chunk_tags[i] for i in result]
savefile(tag_data, result_tags, predict_text)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='VAL')
parser.add_argument('--num', type=int)
parser.add_argument('--embed', type=int)
parser.add_argument('--units', type=int)
parser.add_argument('--gpu', type=int)
args = parser.parse_args()
gpu_config(args.gpu)
# load model
model_dir = 'expr/' + str(args.num) + '/model.h5'
model, (word2idx, chunk_tags) = bilstm_crf_model.create_model(args.embed,
args.units,
train=False)
model.load_weights(model_dir)
test_dir = 'data/raw/test/'
submit_dir = 'data/raw/submit/'
#test(test_dir, submit_dir, model,word2idx, chunk_tags)
txt_data = 'data/raw/local_test/152_6.txt'
tag_data = 'data/raw/local_test/152_6.ann'
local_test(txt_data, tag_data, model, word2idx, chunk_tags)