search

nmichlo / disent

🧶 Modular VAE disentanglement framework for python built with PyTorch Lightning ▸ Including metrics and datasets ▸ With strongly supervised, weakly supervised and unsupervised methods ▸ Easily configured and run with Hydra config ▸ Inspired by disentanglement_lib
https://disent.michlo.dev
MIT License
122 stars 18 forks source link

Standardise Loss Reduction Modes #18

Open nmichlo opened 3 years ago

nmichlo commented 3 years ago

Removed reduction="sum" in #12, but considerable effort is still required to maintain and verify models (#3, #4, #5)

Two options exist:

  1. Allow models to specify their allowed modes: "mean" or "mean_sum"
  2. Only use "mean_sum"
nmichlo commented 2 years ago

"Loss reduction mode" should maybe be called "beta scaling" and be a boolean value?

The problem with the loss reduction modes is that not all the frameworks supports them.