dartsort

⚠️ Work in progress code repository

We do not currently recommend DARTsort for production spike sorting purposes. We are in the process of implementing a robust and documented pipeline in src/dartsort, and we will update this page accordingly.

A workflow described in our preprint (https://www.biorxiv.org/content/10.1101/2023.08.11.553023v1) is in uhd_pipeline.py, which is implemented using the legacy code in src/spike_psvae.

Suggested install steps:

If you don't already have Python and PyTorch 2 installed, we recommend doing this with the Miniforge distribution of conda. You can find info and installers for your platform at Miniforge's GitHub repository. After installing Miniforge, conda will be available on your computer for installing Python packages, as well as the newer and faster conda replacement tool mamba. We recommend using mamba instead of conda below, since the installation tends to be a lot faster with mamba.

To install DARTsort, first clone this GitHub repository.

After cloning the repository, create and activate the mamba/conda environment from the configuration file provided as follows:

$ mamba env create -f environment.yml
$ mamba activate dartsort

Next, visit https://pytorch.org/get-started/locally/ and follow the PyTorch install instructions for your specific OS and hardware needs. We also need to install linear_operator from the gpytorch channel. For example, on a Linux workstation or cluster with NVIDIA GPUs available, one might use (dropping in mamba for conda commands):

# Example -- see https://pytorch.org/get-started/locally/ to find your platform's command.
(dartsort) $ mamba install pytorch torchvision torchaudio pytorch-cuda=11.8 linear_operator -c pytorch -c nvidia -c gpytorch

Finally, install the remaining pip dependencies and dartsort itself:

(dartsort) $ pip install -r requirements-full.txt
(dartsort) $ pip install -e .

To enable DARTsort's default motion correction algorithm DREDge, clone its GitHub repository, and then cd dredge/ and install the DREDge package with pip install -e ..

Soon we will have a package on PyPI so that these last steps will be just a pip install dartsort.

To make sure everything is working:

$ (dartsort) pytest tests/*