def main():
config = args.parser_args()
if os.path.exists(os.path.join(config.embedding_dir,config.embedding_file)):
pretrain_embedding = get_embedding(config.embedding_dir, config.embedding_file)
config.word_embedding=pretrain_embedding
if not os.path.exists(config.save_path):
os.makedirs(config.save_path)
if not os.path.exists(config.restore_path):
os.makedirs(config.restore_path)
if not os.path.exists(config.summary):
os.makedirs(config.summary)
config.train=True
if config.train:
train_file=os.path.join(config.data_dir,config.train_file)
print(train_file)
traindata = Dataprocess(train_file, config.word_dict_file,use_pos_fea=False)
train_data=traindata.get_processdata()
print(type(train_data))
print(len(train_data))
data_train = BatchGenerator(train_data,shuffle=True,maxsen_len=config.max_seq_len)
config.word_vocab_size = traindata.get_vocab_size()
dev_file=os.path.join(config.data_dir,config.dev_file)
devdata = Dataprocess(dev_file, config.word_dict_file,use_pos_fea=False)
dev_data = devdata.get_processdata()
data_dev = BatchGenerator(dev_data,shuffle=False,maxsen_len=config.max_seq_len)
train(config, data_train, data_dev)
s,
output_num, [filter_width],
padding='SAME',
data_format='NWC',
activation_fn=tf.nn.relu,
weights_initializer=xavier_initializer_conv2d(),
weights_regularizer=l2_regularizer(self.l2_reg))
def pooling(self, s, pooling_width):
#一维平均池化只有tf>1.4以上版本有。
'''
:param s:[bs,len,hidden]
:param pooling_width: pooling width
:return:
'''
s = tf.expand_dims(s, 1)
pool = tf.nn.avg_pool(s, [1, 1, pooling_width, 1], [1, 1, 1, 1],
padding='SAME')
return tf.squeeze(pool, axis=1)
if __name__ == '__main__':
import config
config = config.parser_args()
config.word_vocab_size = 5
config.char_vocab_size = 6
config.is_train = True
model = ABCNN(config)