Tropical Cyclone Model

This repository packages the model created in for the EMPIRIC TC Project. This package was created to setup the model to run as a Google Cloud function, and can be installed through pip at the following URL: git+https://github.com/brown-ccv/tc-model.git.

The model requires 2 static data files to be able to run.

1. The model weights (src/tc_model/data/weights.keras)
2. The input data (src/tc_model/data/input_data.hdf5)

The weights file is only 4 MB, and as a result can be included with the repository. The input data is much larger (~100 MB), and has to be hosted separately. Right now, it is being stored intermediately on Google Drive, and downloaded once the tc_model package is installed. Because the Cloud Function will be updated to generate the required input data based on user input parameters, it's not critical to develop a better long term solution to this. When this file is no longer needed, pyproject.toml should be updated to remove pipx as a dependency and the hatch build hook.

If the model is ever retrained, the weights.keras file can be replaced with the new version. Code changes should not be required, unless the original model code was sufficiently updated so that the .keras file can no longer be loaded.

Getting Started

To get started with the code, create a virtual environment and install the dependencies:

$ python3 -m venv .venv
$ source .venv/bin/activate
(.venv) $ pip install --editable

This will install the package from the pyproject.toml in the virual environment. Changes made to the files in src/tc_model will impact the virtual environment.

Editing the Codebase

All files added in src/tc_model will be exposed in the packaged version of the model. For example, a new file tc_model/foo.py with a function bar could be imported like so:

from tc_model.foo import bar

The "API" functions (streamlined consumer facing functionality) is contained within __init__.py, and is imported from tc_model. Adding a new function bar here would be imported like so:

from tc_model import bar

Any new consumer facing function written should have a test associated with it to ensure it returns the correct shape of data for a given set of inputs.

Tests

The tests for this package ensure the functions work correctly. They do not test for accuracy of the model, only that the inputs and outputs are in the expected formats.

To run the tests, set up a virtual environment and install the package with the test configuration:

(.venv) $ pip install --editable ".[test]"

Then, run the tests by running pytest.