Setup

1. Install the package

To get started, follow the installation guide to install the package and all required dependencies.

Tip

If you plan to modify the codebase, it's recommended to fork the repository first. You’ll find relevant setup steps in the contributing section.

2. Download pretrained models

The following script downloads four deterministic models (archesweather-m-seed*) and one generative model (archesweathergen) from Hugging Face:

src="https://huggingface.co/gcouairon/ArchesWeather/resolve/main"
MODELS=("archesweather-m-seed0" "archesweather-m-seed1" "archesweather-m-skip-seed0" "archesweather-m-skip-seed1" "archesweathergen")

for MOD in "${MODELS[@]}"; do
    mkdir -p "modelstore/$MOD/checkpoints"
    wget -O "modelstore/$MOD/checkpoints/checkpoint.ckpt" "$src/${MOD}_checkpoint.ckpt"
    wget -O "modelstore/$MOD/config.yaml" "$src/${MOD}_config.yaml"
done

You can then follow the notebook tutorial to load the models and run inference. For training, refer to the train section.

3. Download ERA5 quantile statistics

ERA5 quantiles are required to compute Brier scores and are used during both inference and training. Download them with:

src="https://huggingface.co/gcouairon/ArchesWeather/resolve/main"
wget -O geoarches/stats/era5-quantiles-2016_2022.nc $src/era5-quantiles-2016_2022.nc