ReLERNN v2.0.0

Major release modernizing the TensorFlow/CUDA stack and build system.

What's new

• TensorFlow 2.19 + Keras 3: Replaces TF 2.15. Fixes GPU underutilization issues (#65, #73) by replacing the broken tf.compat.v1 GPU configuration with native TF2 APIs. Model serialization updated from JSON+H5 to Keras 3 .keras format.
• pixi for dependency management: Single pixi install handles Python, TensorFlow, CUDA, cuDNN, and all dependencies. No more manual conda/pip setup. Supports GPU (default) and CPU-only (cpu) environments.
• Forward-compatible CUDA via cuda-compat: TF 2.19 (CUDA 12.8) works on older NVIDIA drivers (e.g. 535) without driver upgrades.
• Unit tests + CI: 63 tests covering core utility functions, data preprocessing, and simulation methods. GitHub Actions runs tests on every push and PR.
• Python 3.10-3.12 support: Fixes random.seed() with numpy integers (Python 3.12+), NearestNeighbors API change (scikit-learn), Keras 3 model.fit() and plotResults compatibility.

Installation

git clone https://github.com/kr-colab/ReLERNN.git
cd ReLERNN
pixi install

See the README for details.

Breaking changes

• setup.py and requirements.txt removed in favor of pyproject.toml and pixi.toml
• Trained models are now saved as .keras files (not model.json + weights.h5). Models trained with v1.x cannot be loaded directly.
• Requires Python >= 3.10

ReLERNN v1.0.1

Maintenance release. Aligns the package version in setup.py with the GitHub release versioning scheme.

No functional changes from v1.0.0. A major release (v2.0) with TensorFlow 2.19, Keras 3, pixi-based installation, and unit tests is in progress on the tf-migration branch.

ReLERNN v1.0.0

This is the first release of ReLERNN. Much, though not all, of this code was in place for the analyses conducted in our first manuscript (https://doi.org/10.1093/molbev/msaa038). The "releases" function of GitHub will be used from now on to track notable updates to the functionality of ReLERNN.