AthenaEPI / dmipy

The open source toolbox for reproducible diffusion MRI-based microstructure estimation
MIT License
96 stars 30 forks source link

M2 Mac distribtuion #124

Open mrgrist opened 1 year ago

mrgrist commented 1 year ago

Hi there,

The numpy and scipy versions needed for DMIPY are incompatible for an M2 Mac - are there plans to update the numpy / scipy versions to later releases?

36000 commented 1 year ago

I have a similar issue, and would be willing to help update the code if needed

sungoku commented 8 months ago

Is there any progress for this?

mrgrist commented 8 months ago

There was an issue raised by @matteofrigo a while back about the optimisers for SciPy >1.4. @matteofrigo are you able to expand on the behaviour you saw?

mrgrist commented 7 months ago

I have tried compiling with a later version of scipy & fit a data set for NODDI. The results seem ok, but there are a load of numba depreciation warnings. @matteofrigo is this what you saw? Or was it something else? @rutgerfick any thoughts?

mrgrist commented 7 months ago

Ok I have done a benchmarking vs a stable release on an M1 computer.

Average time per voxel on M2 (latest version of scipy) is 1.3s, average time per voxel on M1 (recommended version of scipy) is 0.05s.

This is going to be a problem..