sparkles is a deep learning model that predicts priors for blinx.
For a CPU installation:
conda create -n sparkles python pytorch cpuonly numpy matplotlib zarr jax -c pytorch
git clone https://github.com/funkelab/blinx_sparkle_stats.git sparkles
cd sparkles
pip install .For a GPU installation:
conda create -n sparkles_gpu python pytorch pytorch-cuda cudatoolkit cudatoolkit-dev cudnn jaxlib=0.4.31=*cuda* jax=0.4.31 cuda-nvcc numpy matplotlib zarr -c pytorch -c nvidia
git clone https://github.com/funkelab/blinx_sparkle_stats.git sparkles
cd sparkles
pip install .Note that this installation pins JAX to 0.4.31. You might need to pin certain cuda or jax versions to ensure compatability.
See the scripts directory for example scripts regarding dataset creation, model training, and prediction.