qipengguo / graphwriter-dgl Goto Github PK

把对话中的上一句话作为title输入，知识三元组作为graph输入，作者大大觉得效果会如何呢？

I was learning your code and here is my question about the BucketSampler:

utils.py
mlen = max([0]+[lens[x] for x in batch])
maybe lens should be reinitialized by
lens = torch.Tensor([len(x) for x in self.data_source])
rather than
lens = lens[perm]

if there is anything wrong, please point out and thank you for your help.

我自己也跑了作者的code，感觉生成的质量不算很高。有很多重复且复制的实体不全