search

stsievert / LeanSGD

Wide Residual Networks (WideResNets) in PyTorch
2 stars 2 forks source link

Reshape the layers #5

Open stsievert opened 6 years ago

stsievert commented 6 years ago

Instead of 64x9, maybe 32x18.

This will be _resize_to_2d in codings/svd.py.

stsievert commented 6 years ago

@hwang595

stsievert commented 6 years ago

@hwang implemented this. We looked at the results and decided to reshape (n, 9) to (n/2, 18)