Graph Deep Learning Postprocessing Project @USI

Project designed for the Graph Deep Learning course at Università della Svizzera italiana (USI), focusing on postprocessing wind NWP forecasts.

Quickstart

1. Install the dependencies:

poetry install

2. Activate the environment:

cd spatiotemporal_postprocessing
poetry shell

3. Train

Define the folder with the training data:

export DATA_BASE_FOLDER=<FOLDER>

Define the MLFlow tracking URI (defaults to a local folder called mlruns):

export MLFLOW_TRACKING_URI=<URI>

Train with default settings:

python train.py

Train with a different config:

python train.py --config-name <CFG>

Overwrite (if existing) or append (if not existing) a config value, such as the optimizer:

python train.py ++training.optim.algo=SDG

4. Check the logs on MLflow:

mlflow ui --port <PORT>