problem in data loading

[cgc-cell]

In the line of datasets/shha.py:

        dotseq = torch.from_numpy(np.load(self.dotpath.format(smpid)))[:, :2] * resize_factor

seq is being sliced to include only the first two columns (:2).

Then, in the line of datasets/utils.py

        for i, seq in enumerate(dotseqs):
            u = torch.amin(seq[:, 2:], dim=1, keepdim=True)

you are trying to access seq[:, 2:], i.e., columns starting from index 2.

The u can't get from the dotseq. Is that a bug？

[LrHITer]

+1

[Elin24]

Sry for the late reply.

u is the distance between the point and its nearest point, but in this code, it is automatically calculated in the loss, so it is not used. This part can be deleted:

for i, seq in enumerate(dotseqs):
    dotseqs[i] = seq[:, [1, 0]]

[Elin24]

I have modified the corresponding part of the code. You can check the newest code.