xiaomage

> Hi，I run your code with 'python ./model/Twitter/BiGCN_Twitter.py Twitter15 10' to get the average experimental results of 10 iterations of BiGCN model on Twitter15 (running 100 iterations takes too much...

> I think this [line](https://github.com/HobbitLong/SupContrast/blob/master/losses.py#L89) does it. log_prob = logits - torch.log(exp_logits.sum(1, keepdim=True)) The `logits` denote all samples' distances and `torch.log(exp_logits.sum(1, keepdim=True))` denote the negative samples' distances . The `log_prob...

> I think this [line](https://github.com/HobbitLong/SupContrast/blob/master/losses.py#L89) does it. I still don't understand it. Please help me, thanks.

> > I think this [line](https://github.com/HobbitLong/SupContrast/blob/master/losses.py#L89) does it. > > I still don't understand it. Please help me, thanks. he `log_prob ` denotes all samples' distances and the `mask *...

> 是有batch个图，但序号不是[0,batch-1]吧 [https://pytorch-geometric.readthedocs.io/en/latest/notes/introduction.html#mini-batches](https://pytorch-geometric.readthedocs.io/en/latest/notes/introduction.html#mini-batches) 

>  为啥打印 `batch-data`,？ ` data.x`的组织形式 是【0,0,1,1，.....】 ` rootindex = data.rootindex` 保存的应该是每张图的根节点索引。 比如说，第一张图的根节点是第3个图节点，第二张图的根节点是第10个图节点，`data.rootindex`应该是【3,10,...】以此类推。` x1[rootindex[num_batch]]` 这里是取出来当前一张图的根节点的索引，然后在`x1`中检索。 在这里我产生疑惑是： 当按照10去x1检索时，取出来的不是第二个图的第10个节点，而是从第一张图开始累积的第10个节点. 不知道我有没有描述清楚我的疑惑？ 我并没有打印以上代码，我只是按照作者的代码逻辑，看了一下。

>  如果你运行起来了代码，可以打印一下。也帮助我解决一下我的疑惑，谢谢！

已解决。