FluxML/Flux.jl

Weights shape not validated against kernel, channels

Fermée

#2 506 ouverte le 25 oct. 2024

 (5 commentaires) (1 réaction) (0 personne assignée)Julia (619 forks)batch import
good first issuehelp wanted

Métriques du dépôt

Stars
 (4 725 étoiles)
Métriques de merge PR
 (Métriques PR en attente)

Description

weights = Flux.kaiming_normal()(3, 3, 1)
Conv((3, 3), 1 => 1; pad = (1, 1), init = (_...) -> weights)
# Conv((3,), 3 => 1, pad=1)  # 10 parameters

weights = Flux.kaiming_normal()(3, 3, 1, 1)
Conv((3, 3), 1 => 1; pad = (1, 1), init = (_...) -> weights)
# Conv((3, 3), 1 => 1, pad=1)  # 10 parameters

I wanted to strictly specify the weight init for testing, but encountered this odd result. I think there should be validation to ensure that the weight shape matches the kernel size and input channels, and error if there is a mismatch.

Guide contributeur