Closed GitHunter0 closed 3 years ago
Please try again. I have uploaded windows files for the latest release.
In case of any issues, install numpy+mkl before any other packages
Hey @has2k1 , thanks for the quick feedback!
Now the package installed with no error message.
However, when I try to use stat_smooth(method='loess')
in plotnine
, I receive PlotnineError: "For loess smoothing, install 'scikit-misc'"
As regarding numpy+mkl
, would it be possible to encapsulate it in a package to install with pip install
without the need to download anything? These external dependencies are really problematic when sharing code, so I think it would be great if there is a simpler alternative
scikit-misc
uses c
and fortran
which makes it hard to build binaries for all the OSes. For Linux & OSX, there have a build system running on Github Actions which works well though a pain when it breaks. Windows requires a different build system which I haven't been able to massage into being, I have gotten frustrated the few times I tried.
Otherwise for we rely on those wheels at https://www.lfd.uci.edu/~gohlke/pythonlibs/, which require mkl version of numpy and so the vanilla pip install
wouldn't suffice.
However, when I try to use stat_smooth(method='loess') in plotnine, I receive PlotnineError: "For loess smoothing, install 'scikit-misc'"
If you managed to install numpy+mkl
and scikit-misc
, then I think you installed them in a different environment than the one you are using to plot.
Hi @has2k1 , thank you for the feedback. I understand, it is definitively a very challenging and frustrating task...
My concern is that it is probably a relevant barrier for a more widespread use of the amazing package which is plotnine
.
Have you considered going in other direction and look for simpler alternatives? For example using statsmodels.nonparametric.smoothers_lowess.lowess()
function as the backend for stat_smooth(method='loess')
. That would work out-of-the-box.
If you managed to install numpy+mkl and scikit-misc, then I think you installed them in a different environment than the one you are using to plot.
I did not install numpy+mkl
yet, but I will try it soon
Have you considered going in other direction and look for simpler alternatives? For example using statsmodels.nonparametric.smoothers_lowess.lowess() function as the backend for stat_smooth(method='loess'). That would work out-of-the-box.
Lowess is available and the default if loess is not installed. Note that with lowess you cannot get confidence intervals.
Thanks man! I did not know that was the default behavior
I cannot install
scikit-misc
, I run this in my Conda environment:python -m pip install scikit-misc
And it is throwing a lot of error messages like:Any help would be greatly appreciated.