Romain Brault
Romain Brault
Hi, I don't think the deprecation warnings are a problem. Looks like it's more about OpenMP windows. Try to set the environment variable KMP_AFFINITY=disabled. see: https://github.com/Microsoft/BashOnWindows/issues/785.
@ngoix have you modified this file (fast_dict.pxd)? If not it means that the error comes from scikit itself. @minoriinoue could you tell us your numpy and cython version?
Ok it seems that the error comes from gcc-7 and higher (at the time it has been tested again gcc-6/gcc-5). Note that sklean 0.19 also fixed in the same file...
NB: sklearn 0.20 proposed to simply remove the DTYPE and ITYPE lines instead of the 'import numpy as np'.
Changed to sklearn 0.20 proposed implementation instead of adding the line import numpy as np. It simply removes the DTYPE variables. We need to check how that impacts the rest...
Related issues: https://github.com/scikit-learn/scikit-learn/issues/9272 Possible fix: https://github.com/scikit-learn/scikit-learn/pull/9311
@minoriinoue, thanks for the feedback.
@vinhqdang you can have a look to https://github.com/scikit-learn/scikit-learn/pull/9311/commits/95f1bb04d8d37aa02c0406de3aa7a0f41c61dfcc for the exact fix. Just copy-paste the changes to the file sklearn/neighbors/dist_metrics.pyx. I'll try to do a PR later for this bug....
Any hope to merge this soon @cjolowicz ?
The minimum should be avoiding warnings (#873) and anticipate future breakage(s).