NeuralOperator Developer’s Guide

This guide provides essential information for developers contributing to NeuralOperator.

Installation and Setup

To get started with development, fork the repository on GitHub, then download your fork:

git clone https://github.com/YOURNAME/neuraloperator.git
cd neuraloperator
pip install -e .[dev]

Code Style and Standards

To ensure code clarity and future maintainability, NeuralOperator adheres to a few style principles:

Commenting and docstrings

In general, docstrings use the NumPy format:

def function(arg1: type1, arg2: type2=default)
    """
    Parameters
    ----------
    arg1 : type1
        description of what arg1 'means'
        for the function's behavior
    arg2 : type2, optional
        description arg2
        by default default
    """

For classes, this docstring should go directly below the class declaration:

We also adhere to good in-line commenting practices. When a block’s function is not obvious on its own, add in-line comments with a brief description. For tensor operations, shape annotations are especially helpful where applicable.

Contributing

Follow these steps when making contributions:

Create a new branch for your feature or bug fix:
```
git checkout -b feature-branch
```
Write clean, well-documented code.

3. Add or update tests in appropriate directory. For instance, if your feature adds a model in neuralop/models/mymodel.py, you would add tests to neuralop/models/tests/test_mymodel.py

4. Run the test suite: .. code-block:: bash

pytest neuralop

5. Submit a pull request (PR) on GitHub from your branch to the upstream origin/main. Ensure your PR clearly communicates what you’ve changed or added.

Extending the Library

To add a new neural operator model:

Create a new file in neuralop/models/.
Implement the model as a subclass of BaseModel.

To add a layer:

Create a new file in neuralop/layers
Ensure the layer is a subclass of torch.nn.Module.

Note

For optional bonus points, add an interactive example featuring your new method to ./examples. This helps both us and you: the simpler it is for new users to understand and adapt your method, the more visibility it will get!

Debugging and Troubleshooting

Use torch.set_detect_anomaly(True) for debugging NaN gradients.
Check GPU memory usage with nvidia-smi.
Ensure dependencies are up to date with pip list –outdated.

Contact

For questions or issues, create a GitHub issue or reach out on the discussion forum.