search

decisionintelligence / pathformer

133 stars 16 forks source link

Questions about the inter Attention #1

Open ZiyuFanCSU opened 6 months ago

ZiyuFanCSU commented 6 months ago

Dear author, I have some doubts about the code here: 

        ####inter Attention######
        x = x.unfold(dimension=1, size=self.patch_size, step=self.stride)  # [b x patch_num x nvar x dim x patch_len]
        x = x.permute(0, 2, 1, 3, 4)  # [b x nvar x patch_num x dim x patch_len ]
        b, nvar, patch_num, dim, patch_len = x.shape

        x = torch.reshape(x, (
        x.shape[0] * x.shape[1], x.shape[2], x.shape[3] * x.shape[-1]))  # [b*nvar, patch_num, dim*patch_len]

Could you please explain what nvar means, and why can you use (b x nvar)? Does (b x nvar) represent a new batch?