anthony-zhou/climate_code_conversion

Proof-of-concept: Translating and Modernizing a Climate Model using ChatGPT

Climate Code Conversion

Converting Fortran code in climate models to Python code using ChatGPT.

This work was presented at the Tackling Climate Change with Machine Learning Workshop at NeurIPS 2023.

Access the corresponding workshop paper and poster here: Proof-of-concept: Using ChatGPT to Translate and Modernize an Earth System Model from Fortran to Python/JAX.

Dependencies

Assuming you are running Ubuntu:

1. Install Python 3.
2. pip install -r requirements.txt to install all Python packages
3. pip install -e . to add the local translation package to your path (needed for absolute imports)

Running the CLI

To run the CLI, call python ./translation/main.py. This is a work in progress.

Folder Structure

_codegen, _parsing, and _testing contain independent subroutines for doing those things.
_demos contains Python notebooks demonstrating the results shared in the paper. tests contains unit tests for the translation module. translation contains the translation module, which is primarily a CLI used to translate code automatically.

You can think of translation as a work-in-progress attempt to combine the insights from getting good at _codegen, _parsing, and _testing, even though each can be used individually.

Visualization

Here's a visual representation of the folder structure, courtesy of GitHub Next's repo visualizer: