mne-python: Conda solver fails when installed with Pytorch + Sci-kit

Issue

When I try to install pytorch, sci-kit family of packages, and mne at the same time, conda solver is not able to work through all the dependencies and gets stuck in a recursive solver loop with following error message:

#8 86.88 Collecting package metadata (repodata.json): ...working... done
#8 154.0 Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.
Examining scikit-image:  18%|█▊        | 10/56 [00:20<02:03,  2.69s/it]

Potential Resolution

Potential resolution would be to revise requirements.txt / .yml to be less restrictive. This would allow conda solver to find a solution.

Temporary Workaround

Install MNE and MNE-NIRS with pip instead of conda

About this issue

Original URL
State: closed
Created 3 years ago
Comments: 15 (9 by maintainers)

Most upvoted comments

I can confirm that with mamba the following resolves in about 15 seconds and installation completes in under 90 seconds:

mamba create -n temp
conda activate temp
mamba install mne scikit-image pytorch  # add "-c conda-forge" if it's not your default

drammock on Dec 30, 2021

I also tried installing with mamba, and the install failed due to mayavi dependency:

If you post any more follow-ups, please include the command that led to the output, not just the output.

drammock on Dec 30, 2021

if mne-nirs pins with == a version of pandas I suspect it’s the root cause. Pinning versions with == and not >= makes easily environnements unsolvable.

Message ID: @.***>

agramfort on Dec 30, 2021

I personally recommend pip, but that’s not our official position. If you want to use conda, try mamba – it is much faster and much better in resolving dependencies than conda.

cbrnr on Dec 30, 2021