GammaLearn

GammaLearn is a collaborative project to apply deep learning to the analysis of low-level Imaging Atmospheric Cherenkov Telescopes such as CTA. It provides a framework to easily train and apply models from a configuration file.

Table of Contents

1. Installation
2. Usage
3. Contributing
4. Cite Us
5. License

Installation

For users

GammaLearn uses uv to manage environments and dependencies.

Quick start (local installation)

Prerequisites

Install uv if needed:

curl -Ls https://astral.sh/uv/install.sh | sh

Then install GammaLearn, selecting the cpu or gpu extra depending on your hardware (this pulls in the matching torch/torchvision build):

uv init && uv add gammalearn --extra cpu

or

uv init && uv add gammalearn --extra gpu

For Developers

GammaLearn uses uv to manage environments and dependencies.

Dependencies are defined in pyproject.toml using optional dependency groups:

• cpu → CPU-only installation (used in CI)
• gpu → GPU-enabled installation
• test → testing dependencies

Torch and torchvision are installed via optional extra dependencies.

Clone the repository:

git clone https://gitlab.in2p3.fr/gammalearn/gammalearn.git
cd gammalearn

Install dependencies using uv:

uv sync --extra cpu

For GPU environments:

uv sync --extra gpu

Run GammaLearn:

uv run gammalearn --help

Run tests:

uv run --locked --extra cpu --group test pytest

Note: If dependencies are modified, update the lock file before committing:

uv lock

Development with Docker (alternative to local uv setup)

The published gammalearn/prod image (used to run experiments) is a minimal multistage build that intentionally does not include uv and installs gammalearn non-editably, so it isn't suitable for development. Instead, build a development container from the builder stage of docker/Dockerfile, which still has uv, Python, and the build tools (git, gcc, g++) needed to compile some dependencies, and mount your working copy over it so gammalearn is installed in editable mode from your local sources.

Clone gammalearn locally and go in the gammalearn directory:

git clone https://gitlab.in2p3.fr/gammalearn/gammalearn.git && cd gammalearn

Usage with an IDE (vscode example)

Many IDE's offer the possibility to build and interact with development containers. In VScode, this is handled by the "dev container" extension included in the remote development extension pack (see the extension documentation). The extension can build a container from a Dockerfile and automatically install a vs-code server inside it, allowing to transparently develop the software while using the environment from inside the container. Add the following as .devcontainer/devcontainer.json: it builds the builder stage of docker/Dockerfile, mounts your working copy at /gammalearn (the same path the image already uses as its WORKDIR), and installs gammalearn plus the dev dependency group (lint, test, doc) in editable mode from those mounted sources. As examples, the python extension and ruff linter are installed in the vs-code server running in the container.

// For format details, see https://aka.ms/devcontainer.json
{
	"build": {
		"dockerfile": "../docker/Dockerfile",
		"target": "builder"
	},
	"workspaceMount": "source=${localWorkspaceFolder},target=/gammalearn,type=bind",
	"workspaceFolder": "/gammalearn",
	"postCreateCommand": "uv sync --extra cpu --group dev",
	"postAttachCommand": "uv run gammalearn --help",
	"runArgs": [ "--network=host"],
	"customizations": {
		// Configure properties specific to VS Code.
		"vscode": {
			// Add the IDs of extensions you want installed when the container is created.
			"extensions": [
				"charliermarsh.ruff",
				"ms-python.python",
				"ms-python.vscode-pylance",
				"njpwerner.autodocstring",
				"tamasfe.even-better-toml",
				"wmaurer.change-case"
			]
		}
	}
}

Use --extra gpu instead of --extra cpu in postCreateCommand if you need GPU support in the container (cpu and gpu are declared as mutually exclusive extras, so pass only one).

Note: By default, the dev containers extension will not rebuild the image unless you explicitly rebuild the container in vscode, so pull the latest changes to docker/Dockerfile and rebuild if it has changed.

Once the container is running, you can enter it from another external terminal with

# Get the container ID of your dev container started with vscode
docker ps
# Get a shell in the container
docker exec -it -w /gammalearn container_ID bash

Usage

Run an experiment (Production)

We recommend the use of apptainer. To get the production image of the version of gammalearn you want to use, for instance to get gammalearn v0.13.0, use apptainer pull:

apptainer pull docker://gitlab-registry.in2p3.fr/gammalearn/gammalearn:v0.13.0

This will create a .sif container file that contains a ready to use gammalearn installation. Warning: apptainer can use several GB of disk space as cache when building the .sif file. By default, the cache is located in your home folder ~/.apptainer/cache. You can change this location by setting the APPTAINER_CACHEDIR environment variable. Clean apptainer's cache with apptainer cache clean

You can now run gammalearn from the container to test it:

apptainer run path_to_your_sif_file.sif bash -c "source /gammalearn/.venv/bin/activate && gammalearn --help"

You can run an experiment using apptainer run. Since apptainer containers are read-only by default, you will need to mount the paths to your input and output files. To use nvidia gpus, you will need to specify the --nv option as well. A typical command example:

# Run the experiment in the container
# Parameters:
# --nv                  to use nvidia gpus from inside the container
# CUDA_VISIBLE_DEVICES  env variable used by pytorch to discover the gpus
# NUMBA_CACHE_DIR       a writable directory where numba can store its compiled functions
#                       (needs to be outside of the container, which is read-only)
# CTAPIPE_CACHE         ctapipe needs a writable place, to store its downloaded files.
# Mounts: input (data and settings file) and output directories
#
# We source the venv's activate script (the venv is at /gammalearn/.venv in the image) rather than
# relying on the image's PATH or on `uv run`, because: the published image does not include `uv` at
# all (it's stripped in the final build stage to save space), and apptainer does not reliably apply
# the image's declared PATH for a new user (every user is "new" under apptainer, since the container
# user is always mapped to the host user rather than a user defined in the image).
apptainer run \
    --nv \
    --env "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES" \
    --env "NUMBA_CACHE_DIR=/tmp/NUMBA" \
    --env "CTAPIPE_CACHE=/tmp/CTAPIPE" \
    --mount type=bind,source=/path/to/input/data_dir/,destination=/corresponding/path/in/container/ \
    --mount type=bind,source=/path/to/output/data_dir/,destination=/corresponding/path/in/container/ \
     path_to_your_sif_file.sif bash -c "source /gammalearn/.venv/bin/activate && gammalearn path_to_your_experiment_settings.py"

You can find examples of setting file in the examples and some sample data in example data

Contributing

Contributions are very much welcome: please see CONTRIBUTING.

Cite Us

Please cite

Jacquemont M, Vuillaume T, Benoit A, Maurin G, Lambert P, Lamanna G, Brill A. GammaLearn: A Deep Learning Framework for IACT Data. In36th International Cosmic Ray Conference (ICRC2019) 2019 Jul (Vol. 36, p. 705). DOI: https://doi.org/10.22323/1.358.0705

For reproducibility purposes, please also cite the exact version of GammaLearn you used by citing the corresponding DOI on Zenodo:

License

GammaLearn is distributed under an MIT license.

## Back to top