Issue with Example 'Fig2Replication'

[damianoazzolini]

Hi,
I'm trying to run the example 'Fig2Replication', but I get some errors in the part 'Fit model'.
The first is TypeError: 'module' object is not callable that comes from line 119 of disc_utils.py.
This can be fixed (I suppose) by replacing torch.fft with torch.fft.fft (also at line 120).
After that, however, there is an error at line 121: IndexError: index 1 is out of bounds for dimension 1 with size 1.

Any idea how to fix it?
Thanks

[TrentBrick]

Hi thanks for your issue and sorry for the slow reply!

What version of Pytorch are you using? It seems like a package issue. If you try PyTorch 1.2.0 then does it work?

[damianoazzolini]

I get:

>>> import torch; print(torch.__version__)
1.9.0+cu102

so 1.9. Currently I cannot downgrade to version 1.2.0...

[TrentBrick]

I'm pretty sure it's a package issue. I haven't worked with FFT in a while and am not sure how to fix your particular bug. Apologies! If you do get a solution consider posting it here for others to find!

[damianoazzolini]

I think there is indeed some package issues due to version changing: pytorch/pytorch#49637

[TrentBrick]

Well spotted and hopefully this gets resolved for future compatibility

[juancamilog]

This is an easy fix: the api is greatly simplified since complex numbers are supported out of the box (dtype complex64). one_hot_add becomes torch.fft.ifft(torch.fft.fft(inputs) * torch.fft.fft(shift)).real, which is pretty much the same as in the edward implementation

[TrentBrick]

@juancamilog thanks for sharing this! Going to close this but will point to your answer for anyone using 1.9 or higher.