'conda update -all' with mkl 11.3.1-0 and numpy 1.10.4

[wikiped]

After updating numpy to 1.10.4 mkl is getting installed automatically (if you let it of course). But after installing mkl (latest version 11.3.1-0) running conda update --all offers to downgrade all numpy dependant packages:

The following packages will be downloaded:

package                    |            build
---------------------------|-----------------
numpy-1.9.3                |          py27_p0         2.8 MB  defaults
bottleneck-1.0.0           |       np19py27_0         142 KB  defaults
h5py-2.5.0                 |       np19py27_2         1.4 MB  defaults
numexpr-2.4.4              |      np19py27_p0         137 KB  defaults
scipy-0.16.0               |      np19py27_p0        83.7 MB  defaults
matplotlib-1.4.3           |       np19py27_3        40.8 MB  defaults
numba-0.23.1               |       np19py27_0         1.5 MB  defaults
pandas-0.17.0              |       np19py27_0         5.8 MB  defaults
pymc-2.3.5                 |       np19py27_0         4.4 MB  defaults
pysal-1.10.0               |       np19py27_0        11.1 MB  defaults
pytables-3.2.2             |       np19py27_0         1.4 MB  defaults
scikit-learn-0.16.1        |      np19py27_p0         3.2 MB  defaults
basemap-1.0.7              |       np19py27_0       120.5 MB  defaults
scikit-image-0.11.3        |       np19py27_0        15.8 MB  defaults
statsmodels-0.6.1          |       np19py27_0         4.6 MB  defaults
------------------------------------------------------------
                                       Total:       297.5 MB

The following packages will be UPDATED:

basemap:      1.0.7-np110py27_0  defaults --> 1.0.7-np19py27_0   defaults
bottleneck:   1.0.0-np110py27_0  defaults --> 1.0.0-np19py27_0   defaults
numba:        0.23.1-np110py27_0 defaults --> 0.23.1-np19py27_0  defaults
pysal:        1.10.0-np110py27_0 defaults --> 1.10.0-np19py27_0  defaults
scikit-image: 0.11.3-np110py27_0 defaults --> 0.11.3-np19py27_0  defaults
statsmodels:  0.6.1-np110py27_0  defaults --> 0.6.1-np19py27_0   defaults

The following packages will be DOWNGRADED:

h5py:         2.5.0-np110py27_4  defaults --> 2.5.0-np19py27_2   defaults
matplotlib:   1.5.1-np110py27_0  defaults --> 1.4.3-np19py27_3   defaults
numexpr:      2.4.6-np110py27_1  defaults --> 2.4.4-np19py27_p0  defaults [mkl]
numpy:        1.10.4-py27_0      defaults --> 1.9.3-py27_p0      defaults [mkl]
pandas:       0.17.1-np110py27_0 defaults --> 0.17.0-np19py27_0  defaults
pymc:         2.3.6-np110py27_0  defaults --> 2.3.5-np19py27_0   defaults
pytables:     3.2.2-np110py27_1  defaults --> 3.2.2-np19py27_0   defaults
scikit-learn: 0.17-np110py27_2   defaults --> 0.16.1-np19py27_p0 defaults [mkl]
scipy:        0.17.0-np110py27_0 defaults --> 0.16.0-np19py27_p0 defaults [mkl]

Since mkl package version has no obvious declaration of numpy=1.9 dependance (like np19py27_0) this comes as a surprise and not an easy thing to pin-point when trying to figure out what drives this behavior (in case mkl was installed before numpy update to 1.10.4).

So is mkl actually not compatible with numpy 1.10.4?

Current conda install:

             platform : win-64
        conda version : 3.19.1
  conda-build version : 1.19.0
       python version : 2.7.11.final.0
     requests version : 2.9.1
     root environment : D:\Anaconda  (writable)
  default environment : D:\Anaconda
     envs directories : D:\Anaconda\envs
        package cache : D:\Anaconda\pkgs
         channel URLs : https://repo.continuum.io/pkgs/free/win-64/
                        https://repo.continuum.io/pkgs/free/noarch/
                        https://repo.continuum.io/pkgs/pro/win-64/
                        https://repo.continuum.io/pkgs/pro/noarch/
          config file : D:\Anaconda\.condarc
    is foreign system : False

[aoeftiger]

I encountered the same problem. And I just found a solution: http://stackoverflow.com/questions/35392996/library-not-loaded-libmkl-intel-lp64-dylib-using-anaconda-on-os-x

By uninstalling the mkl package and subsequently deleting the mkl directory (which got created on the conda update), I was able to properly reinstall the mkl package. Importing numpy worked out of the box then.

[wikiped]

Thank you for the link. Unfortunately this is not helping in my situation (although I tried removing mkl folder manually after uninstalling mkl). I actually don't have problems with using numpy - it's just when you run conda update --all you get all those false downgrade warnings, making update --all option useless.

[wikiped]

Looks like this has been fixed with conda >= 4.0

[dprada]

@wikiped Thanks.

Updating conda to 4.0 solved this issue and also the problem of libgfortran #2177 and ContinuumIO/anaconda-issues#686.

I had the following message importing scipy.stats:

ImportError: libgfortran.so.1: cannot open shared object file: No such file or directory

and also:

ImportError: .../lib/python2.7/site-packages/scipy/special/../../../../libgfortran.so.3: version `GFORTRAN_1.4' not found (required by /usr/lib/liblapack.so.3gf)

In my case this was solved Updating conda to 4.0. After that numpy and scypi were also updated to:

    numexpr:      2.4.4-np110py27_p0  [mkl] --> 2.5-np110py27_0   
    numpy:        1.10.2-py27_p0      [mkl] --> 1.10.4-py27_1     
    pytables:     3.2.2-np110py27_0   --> 3.2.2-np110py27_1 
    scikit-learn: 0.17-np110py27_p1   [mkl] --> 0.17.1-np110py27_0
    scipy:        0.16.1-np110py27_p0 [mkl] --> 0.17.0-np110py27_2

[github-actions]

Hi there, thank you for your contribution to Conda!

This issue has been automatically locked since it has not had recent activity after it was closed.

Please open a new issue if needed.