Problems running latest tutorials on NCI

[edoyango]

Hi tennessee,

In testing out Yue's updated pet/2025.06 NCI environment, I ran into a few problems running the tutorials - some related to expecting access to certain things in certain places:

• The cnn training example has an undefined stats_folder variable in it introduced a few weeks ago rename Squish to Squeeze #105.
• in fourcastnext training demo demo, It's unclear where the yaml file in cfg = hydra.compose(config_name="limited_vars_early_stop.yaml") is supposed to come from. Running it as-is just says the file is missing. Check that the config directory '[/opt/conda/envs/pet/lib/python3.11/Training/limited_variables_early_stopping](https://are.nci.org.au/opt/conda/envs/pet/lib/python3.11/Training/limited_variables_early_stopping)' exists and readable
• In the fourcastnext inference notebook, early_stopping is an unrecognized config
• In the bias correction tutorial, I get a different data variable when getting the monthly averaged data from ERA5 - the variable I get is called t2m, but the tutorial shows 2t.
• When running shapefile_trim = pyearthtools.data.transform.region.from_geosearch('Australia') in the region cutting example, I get a write permission error to '/opt/conda/envs/pet/lib/python3.11/site-packages/pyearthtools/data/static/_geographic/Australia/Australia.zip'

[tennlee]

That's fantastic feedback. Thanks so much. Automatically testing the notebooks is unfortunately not very easy, so we rely on manual testing to pick them up. I periodically test them, but obviously a number of errors have crept in. Do you want answers to your questions, or should I just fix everything and let you know when it's done?

[edoyango]

Always happy to help test things!
I'm happy for you to just fix them :)

[tennlee]

Thanks. I have corrected the CNN training tutorial issue that you mentioned. Please let me know if you have any issues running it through from beginning to end.

[tennlee]

• in fourcastnext training demo demo, It's unclear where the yaml file in cfg = hydra.compose(config_name="limited_vars_early_stop.yaml") is supposed to come from. Running it as-is just says the file is missing. Check that the config directory '[/opt/conda/envs/pet/lib/python3.11/Training/limited_variables_early_stopping](https://are.nci.org.au/opt/conda/envs/pet/lib/python3.11/Training/limited_variables_early_stopping)' exists and readable

I am not getting this issue. The config_dir variable is determined dynamically from the location of the fourcastnext.file location, using a relative path to that. Config_dir is then passed to the hydra initialisation.

This could be occurring because I am working with an editable installation rather than purely the modules environment. I would have thought it would work, but there must be a nuance I haven't accounted for. I'll try to find a fix for it shorly, but I expect that may be the cause.

[tennlee]

• In the bias correction tutorial, I get a different data variable when getting the monthly averaged data from ERA5 - the variable I get is called t2m, but the tutorial shows 2t.

I have corrected this issue, thank you.

[edoyango]

Thanks. I have corrected the CNN training tutorial issue that you mentioned. Please let me know if you have any issues running it through from beginning to end.

This is working end-to-end. Bias correction is working too. Thanks!