search

jonbarron / robust_loss_pytorch

A pytorch port of google-research/google-research/robust_loss/
Apache License 2.0
656 stars 88 forks source link

What should be in alpha and scale? #34

Open NeighborhoodCoding opened 2 years ago

NeighborhoodCoding commented 2 years ago

I want to set alpha = 0 and scale = 999 to the lossfun()

but it shold be tensor, to i just coded like, torch.FloatTensor(0) and torch.FloatTensor(999) to alpha and scale.

but there was error like, assert (scale > 0).all()

what is examplary input to the alpha and scale?

thx.

ControllableGeneration commented 1 year ago

According to the gaussian formula, scale being the standard deviation cannot be 0