Towards Optimally Weighted Physics Informed Neural Networks in Ocean Modelling

This repository contains the data sets, source code, and results of the paper. The source code is in the form of Jupyter Notebooks and can be executed by installing both Python 3 and Jupyter Notebooks on any platform. All assets are released under the open source CeCILL license, see LICENSE file.

Install instructions

Requirements

• Python 3.7.3 or newer,
• Git, and
• A Weight and Biases (free) account.

Option 1: Using pip

Install dependecies by running:

pip install -r requirements.txt

and create an instance of jupyter notebooks by:

jupyter notebook

Note: This requirements.txt file was automatically generated from the poetry specs by running poetry export -f requirements.txt --output requirements.txt --without-hashes.

Option 2: Using poetry

This is the preferred option if you plan to do a more extensive experimentation.

Install poetry by running (see instructions if necessary):

curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -

Install project dependencies in a local virtualenv by running in the root of the project:

poetry install

This created a .venv virtual environment with all the dependecies. To active it and run an instance of jupyter notebooks run:

.venv/bin/activate
jupyter notebook

Reproducing results

1. Run jupyter notebook and navigate to Notebooks to find the code used for the paper.
2. You need to enter the WandB key when asked, which you can find under Settings in your WandB account.
3. The Data folder contains the datasets used in the experiments.
4. For the sake of completeness we have incuded in the folder Plots the results and the code to produce the plots used in the paper.