aditya-grover / climate-learn

Source code for ClimateLearn
MIT License
310 stars 49 forks source link

Standardize #63

Closed tung-nd closed 1 year ago

tung-nd commented 1 year ago

An effort to clean and standardize the code:

  1. Instead of having a rollout function and an upsample function, we now have a unified evaluation function used for evaluating models.
  2. Standardize the shape of the prediction to [B, C, H, W]. This is consistent with all metrics and losses.
  3. Remove redundant code in forecasting and downscaling Module classes, and make changes to adhere to changes in model components and metrics.
tung-nd commented 1 year ago

@se0ngbin now we also compute w_mse during evaluation, which can be used for early stopping

jasonjewik commented 1 year ago

If you can't get around to doing these, I'll do it myself sometime in the next couple days.

tung-nd commented 1 year ago

Will really appreciate if you can do it :)

jasonjewik commented 1 year ago

Just checked the notebooks. Seems fine to me. Approved for merge.