Doc tuning parameters - #84
Conversation
|
Hi @SarahAlidoost the best tuning model has ben run against the test dataset. The loss looks okay. Can you review this? Thanks! |
SarahAlidoost
left a comment
There was a problem hiding this comment.
@rogerkuou Thanks! Looks good 👍 I added a few comments. Let me know if something is unclear.
Co-authored-by: SarahAlidoost <55081872+SarahAlidoost@users.noreply.github.com>
|
Hi @SarahAlidoost , I adapted all your suggestions. Do you want to give another look? Thanks! |
| time_features=time_features, | ||
| land_mask=lsm_mask["lsm"], | ||
| patch_size=(1, *spatial_patch_size), | ||
| stride=stride, |
There was a problem hiding this comment.
| stride=stride, | |
| stride=None, |
There was a problem hiding this comment.
We don't need stride for inference. This argument is mainly for creating more data samples during training.
| patch_size[1] * num_patches[0], | ||
| patch_size[2] * num_patches[1], | ||
| ) | ||
| stride = (20, 20) |
There was a problem hiding this comment.
| stride = (20, 20) |
|
|
||
| dataloader_config = DataLoaderConfig( | ||
| batch_size=10, | ||
| shuffle=True, |
There was a problem hiding this comment.
| shuffle=True, | |
| shuffle=False, |
| batch_size=10, | ||
| shuffle=True, | ||
| num_workers=0, | ||
| pin_memory=False, |
There was a problem hiding this comment.
| pin_memory=False, | |
| pin_memory=True, # set it to True when device=cuda |
|
|
||
| source /home/b/b383704/eso4clima/ClimaNet/.venv/bin/activate | ||
|
|
||
| python -u /home/b/b383704/eso4clima/run_best_tuned_model/run_best_tuned_model.py \ |
There was a problem hiding this comment.
| python -u /home/b/b383704/eso4clima/run_best_tuned_model/run_best_tuned_model.py \ | |
| python -u ./Climanet/scripts/run_best_tuned_model.py \ |
|
|
||
| python -u /home/b/b383704/eso4clima/run_best_tuned_model/run_best_tuned_model.py \ | ||
| --experiment-path /work/bd0854/eso4clima/tune/sst_01 \ | ||
| --test-data-dir /work/bd0854/b380103/eso4clima/output/sst/concatenated/ \ |
There was a problem hiding this comment.
should this path point to preprocessed zarr files and not the raw data?
SarahAlidoost
left a comment
There was a problem hiding this comment.
@rogerkuou thanks. A few more comments and then it can be merged.
fix #80
Added a README file documenting the tuning parameters and testing results on the best tuned model.
Tuning testing results are in
/work/<account_id>/eso4clima/tune/run_best_modelTesting loss on year 2022 is 0.0367