Giters

Blealtan / efficient-kan

An efficient pure-PyTorch implementation of Kolmogorov-Arnold Network (KAN).

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

AssertionError

leedrake5 opened this issue · comments

commented

I am pretty sure this is really just a dimensionally issue, but trying to use KANLinear to substitute for nn.Linear to try this approach out. I can use the tutorial to get it to work with MNIST just fine, but it doesn't work well outside the box, almost certainly because I am missing something.

I keep getting the error for forward:

assert x.dim() == 2 and x.size(1) == self.in_features
AssertationError

All I am doing is dropping KANLinear in for nn.Linear, and keeping in_features and out_features the same hidden size. Is there a way forward can be edited to allow non-image inputs?

commented

My fault. I assumed the layer should accept 2d inputs, which is not the case for nn.Linear. The outer dimensions should be simply view as flattened. Will fix soon.