Installation¶
Braincoder requires Python 3.10+ and Keras 3 with at least one backend (TensorFlow, JAX, or PyTorch).
Recommended: conda + editable install¶
This approach gives you a full development environment with TensorFlow as the backend:
git clone https://github.com/Gilles86/braincoder.git
cd braincoder
conda env create -f environment.yml
conda activate braincoder
The environment.yml installs TensorFlow, Keras, and braincoder itself in editable
mode (pip install -e .), so local changes to the source are reflected immediately.
Choosing a backend¶
Keras 3 supports three backends. Set the KERAS_BACKEND environment variable before
importing braincoder:
export KERAS_BACKEND=tensorflow # default
export KERAS_BACKEND=jax
export KERAS_BACKEND=torch
Or set it permanently in ~/.keras/keras.json:
{"backend": "tensorflow"}
Each backend must be installed separately. The environment.yml includes all three
(tensorflow, jax[cpu], torch), but you can install only the one you need.
pip-only install¶
braincoder is not on PyPI, so install a tagged release from GitHub, together with a backend:
pip install "braincoder @ git+https://github.com/Gilles86/braincoder.git@v0.6.2" tensorflow # TensorFlow
# or
pip install "braincoder @ git+https://github.com/Gilles86/braincoder.git@v0.6.2" "jax[cpu]" # JAX
# or
pip install "braincoder @ git+https://github.com/Gilles86/braincoder.git@v0.6.2" torch # PyTorch
Pin a tag like this in your project’s environment, so that its results stay reproducible.
For a development install from source:
git clone https://github.com/Gilles86/braincoder.git
cd braincoder
pip install -e . tensorflow
Additional dependencies¶
Some features require optional packages:
MCMC sampling:
blackjax(JAX backend recommended)Notebooks / tutorials:
jupyter,ipykernelTests:
pytest(orpip install -e ".[test]"from a clone)
Verifying the install¶
python -c "import braincoder; print(braincoder.__version__)"
pytest tests/