Maintenance and Troubleshooting

This page provides guidance on keeping CPFA up to date and resolving typical issues.

Maintenance

Updating the Repository

To update CPFA to the latest version:

Open Anaconda Prompt.
Navigate to the CPFA project folder.
Pull the latest changes from GitHub (if using git directly):
```
git pull origin main
```

Reinstalling Dependencies

If dependencies become inconsistent, you can reinstall them:

conda activate cpfa_env
pip install --upgrade --force-reinstall \
    numpy pandas matplotlib xarray cartopy \
    onnx==1.13.1 onnxruntime==1.14.0

Managing Output Size

Prediction and visualization steps can generate a large number of files. Periodically clean the output_data folder by removing old experiments that are no longer needed.

Troubleshooting

Environment Not Found

Symptom: Conda reports that the environment cannot be found.

Ensure the environment name is correct (for example, cpfa_env).
Run conda env list to see all available environments.

Import Errors

Symptom: ModuleNotFoundError or similar import issues.

Confirm that the correct environment is activated.
Reinstall the required packages.
Verify that the Python version is 3.9.2 as documented.

ONNX Runtime Errors

Symptom: Errors mentioning ONNX or ONNX Runtime when running the prediction script.

Check that ONNX and ONNX Runtime versions match the documented versions:
- onnx==1.13.1
- onnxruntime==1.14.0
Reinstall both packages to ensure consistency.

Incorrect or Missing Input Data

Symptom: The prediction script fails because it cannot find input_surface.npy or because the file has an unexpected shape.

Verify that input_surface.npy exists in the input_data folder.
Confirm that the array shape is (4, 721, 1440) and that the variable order is correct.

Unexpected Prediction Results

Symptom: Predictions look unrealistic or inconsistent.

Check that the ERA5 data used to generate input_surface.npy matches the documented variable order and units.
Confirm that the correct ONNX model file is being used.

Getting Additional Help

If an issue persists:

Open a GitHub issue with:
- A clear problem description
- Steps to reproduce
- Error messages or logs
- Environment details (OS, Python, package versions)
Alternatively, contact the maintainers via the e-mail addresses listed on the project communication channels.