Hi Peter, good idea. As far as I can tell for know it seems to work fine but again no extensive testing done.
There is however 1 issue I encountered with pillow. I think the conda-forge one or a dependency of it is broken. Eg. pillow must be installed from defaults manually and not via dependency of rdkit. If the conda-forge one is used, there is an error of missing dll when displaying molecules in jupyter. So installation order was as below: conda create --name rdkit_forge python=3.7 conda activate rdkit_forge conda install pillow conda install -c conda-forge ipykernel openblas numpy pandas rdkit "libblas=*=*openblas" For scipy and hence also scikit-learn one still needs to use pip as there is only an mkl build available even in conda-forge. Also xgboost will need to come from pip as there is no windows build on conda-forge and the one on defaults isn't compatible with openblas. There are probably more such issues but I guess still better than getting everything but rdkit from pypi. Best Regards, Thomas ________________________________ Von: Peter St. John <peterc.stj...@gmail.com> Gesendet: Dienstag, 12. November 2019 16:25 An: Greg Landrum <greg.land...@gmail.com> Cc: Thomas Strunz <beginn...@hotmail.de>; rdkit-discuss@lists.sourceforge.net <rdkit-discuss@lists.sourceforge.net> Betreff: Re: [Rdkit-discuss] Anaconda installation without hard dependency on Intel MKl (windows) Another option would be to try the conda-forge rdkit. It doesn't appear to use MKL -- I think the MKL dependency for the rdkit::rdkit package is coming from the defaults::numpy dependency. some tools for example scipy and pandas are only available as openblas builds via pypi (pip). I believe the conda-forge recipes for these are also based on openblas. You might try installing rdkit from conda-forge in a fresh environment and see if those numpy / scipy builds work for you. -- Peter On Tue, Nov 12, 2019 at 6:48 AM Greg Landrum <greg.land...@gmail.com<mailto:greg.land...@gmail.com>> wrote: On Tue, Nov 12, 2019 at 2:00 PM Thomas Strunz <beginn...@hotmail.de<mailto:beginn...@hotmail.de>> wrote: So for me this is temporary workaround but not really a permanent long term solution (and as far as I can tell mostly an issue of conda and windows and not rdkit) Yeah, it's clearly not the idea solution to the problem. And, yes, the problem is clearly related to conda and windows. It seems that there used to be a blog post explaining this (linked from this github issue: https://github.com/ContinuumIO/anaconda-issues/issues/656), but the URL no longer works and I can't find it. If the performance difference is that dramatic on AMD CPUs, it may be worth raising a new issue in the repo above and see if you get any kind of response. -greg _______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net<mailto:Rdkit-discuss@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
_______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
