Closed vncntprvst closed 2 years ago
Hi Vincent, this looks a bad installation related to compilers. This yaml file use to work some month ago. I would try to create an empty env and install all list one by one to track the problem.
Hi Samuel, Thanks for the reply. Yes, it's likely an issue with mpi4py, that is conflicting with a previous install (even though I uninstalled it). If anyone gets the same issue, here's a working recipe:
conda create --name si_env python=3.8
conda install pip>=21.0 numpy=1.20.3 joblib tqdm scipy h5py scikit-learn matplotlib pandas networkx
conda install -c conda-forge datalad
pip install PyQt5 ephyviewer neo>=0.10.0 elephant>=0.10.0 probeinterface MEArec>=1.7.1
pip install spikeinterface[full] spikeinterface-gui
pip install tridesclous>=1.6.3 phy==2.0b5
Thank you for the feedboack. Maybe we will remove this mpi from env in windows it is a bit too complicated.
I was about to say that I'll work on creating a containerized version, so that people can run spikeinterface without having to troubleshoot any installation issue, but then I saw that docker containers are already available. I'll play with those first :)
Hi Vincent, we already have container for several sorter. See this:
The best way to avoid this installation nightmare is to install only spikeinterface on the host and then run sorter in docker/singularity. So no more need to install then. But this don't work on windows host.
I checked those, and that's a great resource. As you mentioned, and documented in this issue, spikeinterface can't run those on Windows. That's a clear limitation. Not every lab can have a big GPU computing station and dedicate it to spike sorting, and use Ubuntu for that purpose. Some will opt to have it run on Windows, as this is still the OS most people are familiar with. That said, WSL2 is a really good compromise now for that user case. Alternatively, one could build a container with SI + a few spike sorters, and simply mount the data/current directory as a volume in the container.
We plan soon to make a big container with spikeinetrface + many sorters inside. We would update this lets says every 3 month.
That's great! Looking forward to it. I made a working version in the meantime. See commit here. For Windows users, one can just drag and drop their data folder on the batch file to start Jupyterlab with Spike Interface and a bunch of sorters installed.
When activating the si_env created with
full_spikeinterface_environment_windows.yml
, I get the error below. Removing msmpi and its .dlls then re-creating the environment did not resolve the issue. Visual Studio and C++ Build SKU are installed.