yuanyz0825

It is very interesting that the work of deepsnap is to repeatedly convert other datasets into deepsnap datasets, but it does not provide the reading and saving interface of deepsnap...

I don't know why the configuration files generated by using grid search are different from the initial settings. I didn't enumerate their changes.

torch.topk 函数设置 largest=False 是为了找到最小的距离，即最近的邻居。这种方法可能比转换张量为 numpy 数组并使用 scipy.spatial.cKDTree 更有效，因为它避免了 CPU 和 GPU 之间的数据传输。

1

@staticmethod def _e_list_from_feature_kNN(features: torch.Tensor, k: int): r"""Construct hyperedges from the feature matrix. Each hyperedge in the hypergraph is constructed by the central vertex and its :math:`k-1` neighbor vertices. Args: ``features``...

当处理图级的超图分类问题，如果每个超图的节点和超边数量较多，batch加速处理对邻接矩阵进行类似dgl.batch的拼接方式时，似乎并不支持很大的batch_size，如何优化？或者在计算速度和显存之间取舍？