Containerized Sorters

One of the biggest bottlenecks for users is installing spike sorting software. To alleviate this, we build and maintain containerized versions for most spike sorters.

The containerized approach has several advantages:

  • Installation is much easier.

  • Different spike sorters with conflicting dependencies can be easily run side-by-side.

  • The results of the analysis are more reproducible and not dependant on the operating system

  • MATLAB-based sorters can be run without a MATLAB licence.

The containers can be run in Docker or Singularity, so having Docker or Singularity installed is a prerequisite.

Running sorters in container docker/singularity

Running spike sorting in a docker container container just requires to:

  1. have docker installed

  2. have docker python SDK installed (pip install docker)


  1. have singularity installed

  2. have singularity python (pip install spython)

Some sorters are GPU required or optional. To run containerized sorters with GPU capabilities, CUDA and nvidia-container-toolkit needs to be installed. Only NVIDIA GPUs are supported for now.

For Docker users, you can either install Docker Desktop (recommended for Windows and MacOS) or the Docker Engine (recommended for Linux). To enable Docker Desktop to download the containers, you need to create an account on DockerHub (free) and perform the login in Docker Desktop. For Docker Engine, you also need to enable Docker to run without sudo priviledges following this post-install guide

The containers are built with Docker, but Singularity has an internal mechanism to convert docker images. Using Singularity is often prefered due to its simpler approach with regard to root privilege.

The following code creates a test recording and runs a containerized spike sorter (Kilosort 3):

import spikeinterface.extractors as se
import spikeinterface.sorters as ss
test_recording, _ = se.toy_example(
test_recording ="test-docker-folder")

sorting = ss.run_kilosort3(


This will automatically check if the latest compiled kilosort3 docker image is present on your workstation and if it is not the proper image will be downloaded from SpikeInterface’s Docker Hub repository. The sorter will then run and output the results in the designated folder.

To run in Docker instead of Singularity, use docker_image=True.

sorting = ss.run_kilosort3(recording=test_recording, output_folder="kilosort3", docker_image=True)

To use a specific image, set either docker_image or singularity_image to a string, e.g. singularity_image="spikeinterface/kilosort3-compiled-base:0.1.0".

sorting = ss.run_kilosort3(

NOTE: the toy_example() returns in-memory objects, which are not bound to a file on disk. In order to run spike sorting in a container, the recording object MUST be persistent on disk, so that the container can reload it. The save() function makes the recording persistent on disk, by saving the in-memory test_recording object to a binary file in the test-docker-folder folder.


The containerization of spike sorters is managed on a separate GitHub repo, spikeinterface-dockerfiles. If you find an error with a current container or would like to request a new spike sorter, please submit an Issue to this repo.