ImportError: cannot import name NUMPY_MKL

If you look at the line which is causing the error, you'll see this:

from numpy._distributor_init import NUMPY_MKL  # requires numpy+mkl

This line comment states the dependency as numpy+mkl (numpy with Intel Math Kernel Library). This means that you've installed the numpy by pip, but the scipy was installed by precompiled archive, which expects numpy+mkl.

This problem can be easy solved by installation for numpy+mkl from whl file from here.

Reinstall numpy-1.11.0_XXX.whl (for your Python) from www.lfd.uci.edu/~gohlke/pythonlibs. This file has the same name and version if compare with the variant downloaded by me earlier 29.03.2016, but its size and content differ from old variant. After re-installation error disappeared.

Second option - return back to scipy 0.17.0 from 0.17.1

P.S. I use Windows 64-bit version of Python 3.5.1, so can't guarantee that numpy for Python 2.7 is already corrected.

I'm not sure if this is a good solution but it removed the error. I commented out the line:

from numpy._distributor_init import NUMPY_MKL 

and it worked. Not sure if this will cause other features to break though

I had the same problem while installing gensim on windows. Gensim is dependent on scipy and scipy on numpy. Making all three work is real pain. It took me a lot of time to make all there work on same time.

Solution: If you are using windows make sure you install numpy+mkl instead of just numpy. If you have already installed scipy and numpy, uninstall then using "pip uninstall scipy" and "pip uninstall numpy"

Then download numpy-1.13.1+mkl-cp34-cp34m-win32.whl from http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy and install using pip install numpy-1.13.1+mkl-cp34-cp34m-win32.wh Note: in cp34-cp34m 34 represent the version of python you are using, so download the relevant version.

Now download scipy from http://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy (appropriate version for your python and system) and install using "pip install scipy‑0.19.1‑cp34‑cp34m‑win32.whl"

Your numpy and Scipy both should work now. These binaries by Christoph Gohlke makes it very easy to install python packages on windows. But make sure you download all the dependent packages from there.