Using it on Windows?
hoechenberger opened this issue · 14 comments
Hello, has anyone here tried to use this package on Windows? I'm having a bit of trouble getting PyTorch to work on this platform via conda-forge, as there's currently not conda-forge package of PyTorch available: mne-tools/mne-installers#123 (comment)
Just wanted to ask if anybody's got experience with this sort of issue?
Yes, but I'm not using conda and I'm getting PyTorch from pip. Works like a charm :)
How do you install Python?
Edit: and NumPy :)
On windows, I grab the installer 64 bits here: https://www.python.org/downloads/release/python-3912/ (or a different version), install it for all users (I don't like installing in the user directories), and add the python directories to the system path (through the installer).
Then, since I'm using Spyder as IDE, I usually install it in the base environment with py -m pip install spyder. I create a virtual env for whatever project I'm working on with py -m venv name, and activate it with name/Scripts/activate.bat.
Then, I install the dependencies I need, e.g.:
py -m pip install --upgrade pip setuptools wheel
py -m pip install spyder-kernels matplotlib pyqt5
py -m pip install numpy scipy mne torch
I'm doing very similarly on Linux or macOS where I install python via the package manage or homebrew and repeat the other steps.
The beauty of python, you have 100 different ways of setting it up.. and messing it up! Someone recently shared this nice illustration for macOS:
The comic is outdated, macOS doesn't ship with system Python anymore :)
I'm still running 11.6 Big Sur, too scared to update :)
Glad to know that they finally got rid of this 2.7 default version!
I haven't used a Windows computer in over 5-10(?) years now, so unfortunately I cannot help :p
Yes, same … But I suppose I should get one specifically for this kind of stuff
@hoechenberger I'm not using a windows machine, but if you want to try installing PyTorch on your own and using the source code from git directly, the PyTorch page describes how to install it on windows: https://pytorch.org/get-started/locally/.
Hello, I report that my windows machine can run mne-icalabel on micromamba environment.
The solution is simply to add a pytorch channel.
micromamba create -n hoge python=3.10 mne=1.3 mne-icalabel=0.4 -c conda-forge -c pytorchAdditionally, the installation document (Installation via Cond) includes wrong information.
This command does not install mne-icalabel on windows.
I think you had better update the documentation.
This project is very great for me. Thanks to all contributors. 😄
Hello, thanks for the report. I will try to fix the documentation for conda.
cc @larsoner, we should try that for our Windows standalone MNE installers
@hoechenberger @mscheltienne I opened a pull request that adds support for ONNX on top of PyTorch. I have not tested it personally on windows (outside the GitHub unit-tests), but there is a conda-forge link to the onnxruntime library. We can add additional functionality to choose which underlying library to use when installing mne-icalable if installing PyTorch is a problem.
Closing as this should now be fixed with the ONNX backend and with MNE-installers.
Yes, thank you!