New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Refactor losses #23

Open

4 tasks

vitusbenson opened this issue Oct 28, 2022 · 0 comments

Collaborator

vitusbenson commented Oct 28, 2022

The losses are currently a mess and need a better API.

Goal:

Move dataset & model specific out of the losses
More explicit naming
Docstrings that indicate the maths behind each loss

Requirements:

Need new abstraction that can handle seperation of loss computation from dataset/model specific stuff
First version should support:
- L1 loss
- L2 loss
- masking (both by data quality mask & by landcover mask)
- Weight decay (as a regularization)

Culprits:

Where to move the masking of preds & targets? Currently it is computed inside the loss function. However, it might be very dataset specific, so what is a good API for this?

Risks:

Will likely break backward-compatibility

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment