Skip to content

Installation

There are 2 options to install geoarches: 1. [Recommended for first time users] Install as a package from PyPI if you don't intend to make any modifications to the code. 2. Install from source if you intend to make modifications to the code.

Option 1: Install from PyPI

geoarches supports Python 3.11 through 3.14.0. We recommend installing it in a virtual environment.

pip

python -m venv .venv
source .venv/bin/activate
python -m pip install geoarches

uv

Install uv by following its installation instructions, then run:

uv venv --python 3.12
source .venv/bin/activate
uv pip install geoarches

You can also activate an existing Conda environment and run python -m pip install geoarches in it.

Verify the installation with:

python -c "import geoarches; print(geoarches.__version__)"

Option 2: Install from source

Clone the repository only if you want to contribute to geoarches or use unreleased changes:

git clone https://github.com/INRIA/geoarches.git
cd geoarches

Install uv and run:

uv sync

Alternatively, install Poetry 2.2 or later and run in a virtual environment:

poetry install

Both source-installation methods install geoarches in editable mode and include the development dependencies. See the Contributing Guide for the complete development workflow.

Building the documentation locally

Documentation dependencies are opt-in. Install them with uv sync --group docs or poetry install --with docs from a source checkout.

Useful directories

In the working directory for your project, we recommend creating these directories or symlinks:

ln -s /path/to/data/ data # (1)!
ln -s /path/to/models/ modelstore # (2)!
ln -s /path/to/evaluation/ evalstore # (3)!
ln -s /path/to/wandb/ wandblogs # (4)!
  1. data/: stores all datasets used for training and evaluation.
  2. modelstore/: stores model checkpoints and Hydra configs.
  3. evalstore/: stores intermediate model outputs used for evaluation metrics.
  4. wandblogs/: stores Weights & Biases logs.

You can create regular directories instead. Missing directories are created in the current working directory when needed.

Downloading data

The download/ folder contains scripts to download data. To download the full ERA5 dataset from WeatherBench for training and evaluation, run:

python -m geoarches.download.dl_era --folder /path/to/data/era5_240/full/

If you only want to run evaluations, download the years 2019 to 2021:

python -m geoarches.download.dl_era --folder data/era5_240/full/ --years 2019 2020 2021

Working with ArchesWeather and ArchesWeatherGen

To use ArchesWeather or ArchesWeatherGen, follow the ArchesWeather setup instructions.