You can run Kilosort4 without installing it locally using google colab. An example colab notebook is available here. It will download some test data, run kilosort4 on it, and show some plots.
Example notebooks are provided in the docs/source/tutorials
folder and in the docs. The notebooks include:
basic_example
: sets up run on example data and shows how to modify parametersload_data
: example data format conversion through SpikeInterfacemake_probe
: making a custom probe configuration.
If you use Kilosort1-4, please cite the paper:
Pachitariu, M., Sridhar, S., Pennington, J., & Stringer, C. (2024). Spike sorting with Kilosort4.
Linux and Windows 64-bit are supported for running the code. At least 8GB of GPU RAM is required to run the software (see docs for more recommendations). The software has been tested on Windows 10 and Ubuntu 20.04.
If you have an older kilosort
environment you can remove it with conda env remove -n kilosort
before creating a new one.
- Install an Anaconda distribution of Python. Note you might need to use an anaconda prompt if you did not add anaconda to the path.
- Open an anaconda prompt / command prompt which has
conda
for python 3 in the path - Create a new environment with
conda create --name kilosort python=3.9
. Python 3.10 should work as well. - To activate this new environment, run
conda activate kilosort
- To install kilosort and the GUI, run
python -m pip install kilosort[gui]
. If you're on a zsh server, you may need to use ' ' around the kilosort[gui] call: `python -m pip install 'kilosort[gui]'. - Instead of 5, you can install the minimal version of kilosort with
python -m pip install kilosort
. - Next, if the CPU version of pytorch was installed (will happen on Windows), remove it with
pip uninstall torch
- Then install the GPU version of pytorch
conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia
Note you will always have to run conda activate kilosort
before you run kilosort. If you want to run jupyter notebooks in this environment, then also conda install jupyter
or pip install notebook
, and python -m pip install matplotlib
.
If step 8 does not work, you need to make sure the NVIDIA driver for your GPU is installed (available here). You may also need to install the CUDA libraries for it, we recommend CUDA 11.8.
If pytorch installation still fails, follow the instructions here to determine what version of pytorch to install. The Anaconda install is strongly recommended on Windows, and then choose the CUDA version that is supported by your GPU (newer GPUs may need newer CUDA versions > 10.2). For instance this command will install the 11.8 version on Linux and Windows (note the torchvision
and torchaudio
commands are removed because kilosort doesn't require them):
conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia
This video has step-by-step installation instructions for NVIDIA drivers and pytorch in Windows (ignore the environment creation step with the .yml file, we have an environment already, to activate it use conda activate kilosort
).
- Open the GUI with
python -m kilosort
- Select the path to the binary file and optionally the results directory. We recommend putting the binary file on an SSD for faster processing.
- Select the probe configuration (mat files recommended, they actually exclude off channels unlike prb files)
- Hit
LOAD
. The data should now be visible. - Hit
Run
. This will run the pipeline and output the results in a format compatible with Phy, the most popular spike sorting curating software.
There is a warning that will always pop up when running Kilosort and/or using the BinaryFile class, but it's nothing to worry about:
UserWarning: The given NumPy array is not writable, and PyTorch does not support non-writable tensors. This means writing to this tensor will result in undefined behavior. You may want to copy the array to protect its data or make it writable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at C:\cb\pytorch_1000000000000\work\torch\csrc\utils\tensor_numpy.cpp:205.)
Phy provides a manual clustering interface for refining the results of the algorithm. Kilosort4 automatically sets the "good" units in Phy based on a <10% estimated contamination rate with spikes from other neurons (computed from the refractory period violations relative to expected).
Check out the Phy repo for more install instructions. We recommend installing Phy in its own environment to avoid package conflicts.
After installation, activate your Phy environment and navigate to the results directory from Kilosort4 (by default, a folder named kilosort4
in the same directory as the binary data file) and run:
phy template-gui params.py
To get the most up-to-date changes to the code, clone the repository and install in editable mode in your Kilosort environment, along with the other installation steps mentioned above.
git clone https://github.com/MouseLand/Kilosort.git
conda activate kilosort
pip install -e Kilosort[gui]
pip uninstall torch
conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia
For unit testing, you will need to install pytest
pip install pytest
Then run all tests with:
pytest --runslow
To run on GPU:
pytest --gpu --runslow
Omitting the --runslow
flag will only run the faster unit tests, not the slower regression tests.