Bris inference

This is a package to run MET Norway's data-driven model Bris, which is based on the Anemoi framework.

Features

• Model and data-parallel inference
• Multi encoder/decoder
• Time interpolation
• Ensembles

Documentation

See Wiki

Requirements

• udunits2 library. On ubuntu available as libudunits2-0
• Running on ARM/MacOS requires some workarounds for now: metno#85

Install

Locally for development

python3 -m venv venv && source venv/bin/activate
pip install -e .

From PIP

pip install bris

Via docker, if you are Met.no employee

See Dockerfile

How to run tests

pip install -e '.[dev]'
tox

When pushing to github, default tests will be run automatically and must succeed. Read more about Tests in the wiki.

Code borrowed from Anemoi project

• bris/ddp_strategy.py is based on https://github.com/ecmwf/anemoi-core/blob/main/training/src/anemoi/training/distributed/strategy.py
• bris/grid_indices.py is based on https://github.com/ecmwf/anemoi-core/blob/main/training/src/anemoi/training/data/grid_indices.py
• bris/data/data{set,module}.py is somewhat based on https://github.com/ecmwf/anemoi-core/tree/main/training/src/anemoi/training/data