User Guide
This section provides detailed guidance for using geoarches.
If you’re just getting started, begin with the Getting Started section for installation and basic usage.
Prerequisites
geoarches builds on top of several open-source tools. While not strictly required, familiarity with the following tools will help you get the most out of the library.
Hydra
We use Hydra for flexible and modular configuration of training experiments.
The main entry point is main_hydra.py, which builds the full configuration from components located under the configs/ directory. This includes:
- The base config:
configs/config.yaml - Module-specific configs: e.g.
configs/module/archesweather.yaml - Dataloader configs: e.g.
configs/dataloader/era5.yaml
You can override any argument via the command line (see Pipeline API for the full list).
Example
python -m geoarches.main_hydra \
module=archesweather \ # (1)!
dataloader=era5 \ # (2)!
++name=default_run # (3)!
- Loads
configs/module/archesweather.yaml - Loads
configs/dataloader/era5.yaml - Unique name of your run, used for checkpointing and W&B logging
PyTorch & PyTorch Lightning
We rely on PyTorch Lightning to simplify training and evaluation, removing much of the boilerplate around training loops.
In particular, we use the LightningModule abstraction to wrap backbone models, handle loss computation, optimizer setup, logging, and more.
Note
If you're only interested in the data or evaluation utilities provided by geoarches, you do not need to use Lightning.
Weights & Biases (W&B)
Optionally, you can log training metrics with Weights & Biases. It provides experiment tracking for your runs.