Potential issue with Pytorch MIONet when using input transform

[Dkupetsk]

Hello,

I encountered an error in my project when I tried to apply a feature transform in MIONet:

"RuntimeError: mat1 and mat2 shapes cannot be multiplied (1000x1 and 6x100)"

My feature transform was supposed to send my trunk network input to a MxNx6 array. I noticed that in the MIONet PyTorch the code says the following:

if self._input_transform is not None:
    y_loc = self._input_transform(x_loc)
y_loc = self.trunk(x_loc)

After changing this to

if self._input_transform is not None:
    x_loc = self._input_transform(x_loc)
y_loc = self.trunk(x_loc)

I no longer had any issues. Is this a potential typo?

Thanks

[lululxvi]

The code looks correct.

deepxde/deepxde/nn/pytorch/deeponet.py

Line 52 in 26860b1

if self._input_transform is not None:

Can you point out the code?

[Dkupetsk]

Yeah it looks fine in DeepONet, but MIONet is different:
see here

if self._input_transform is not None:
            y_loc = self._input_transform(x_loc)
        y_loc = self.trunk(x_loc)

shouldn't it say

if self._input_transform is not None:
            x_loc = self._input_transform(x_loc)
        y_loc = self.trunk(x_loc)

[lululxvi]

Yes, you are right. It is a bug. Do you want to submit a PR to fix this?

[Dkupetsk]

Sure!