def demo():
word2int, int2word = read_dict()
tag2label = {'0': 0, '1': 1}
int2tag = {l: t for t, l in tag2label.items()}
model_path = os.path.join(MODEL_PATH, FLAGS.DEMO, 'checkpoints')
ckpt_file = tf.train.latest_checkpoint(model_path)
logger.info("load model from {}".format(ckpt_file))
textCNN = TextCNN(
model_path=ckpt_file,
vocab=word2int,
tag2label=tag2label,
eopches=FLAGS.epoches,
)
saver = tf.compat.v1.train.Saver()
with tf.compat.v1.Session(config=cfg()) as sess:
print('============= demo =============')
saver.restore(sess, ckpt_file)
while True:
print('Please input your sentence:')
inp = input()
if inp == '' or inp.isspace():
print('See you next time!')
break
else:
inps = [inp.strip()]
pred = textCNN.predict(sess, inps)[0]
probs = textCNN.predict_prob(sess, inps)[0]
print("\n{}".format(inps))
for idx, prob in enumerate(probs):
print("\t{} -> {}".format(int2tag[idx], prob))
print("\tTag: {}".format(int2tag[pred]))
parser.add_argument("--decay_rate", type=float, default=0.9, help="the decay rate for lr")
parser.add_argument("--sequence_length", type=int, default=50, help="sequence length")
parser.add_argument("--vocab_size", type=int, default=150346, help="the num of vocabs")
parser.add_argument("--embed_size", type=int, default=200, help="embedding size")
parser.add_argument("--is_training", type=bool, default=True, help='training or not')
parser.add_argument("--keep_prob", type=float, default=0.9, help='keep prob')
parser.add_argument("--clip_gradients", type=float, default=5.0, help='clip gradients')
parser.add_argument("--filter_sizes", type=list, default=[2, 3, 4], help='filter size')
parser.add_argument("--num_filters", type=int, default=128, help='num filters')
parser.add_argument('--mode', type=str, default='train', help='train|test|demo')
parser.add_argument('--DEMO', type=str, default='tf_rnn', help='model for test and demo')
return parser.parse_known_args()
FLAGS, unparsed = args()
word2int, int2word = read_dict()
tag2label = {'0': 0, '1': 1}
int2tag = {l: t for t, l in tag2label.items()}
def cfg():
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 0.2 # need ~700MB GPU memory
return config
def train():
iter = -1
iter_size = 20000
train, dev = read_corpus(random_state=1234, separator='\t', iter=iter, iter_size=iter_size)