11

I've already installed numpy 1.9.0 in Python.Now what should I do to get numpy+mkl?

Hari
  • 357
  • 2
  • 3
  • 12
  • If the 1.9 is not a strict requirement, and higher version allowed, I would suggest downloading the wheel from [here](http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy) and install it using `pip` for example. For me - it was the simplest way to install all needed packages on x64 win 7 without setting up additional things like Anaconda. – Mikhail Churbanov Dec 19 '16 at 07:48

3 Answers3

23

If you do not have an entire Python distribution or you do not want to install one, you can download and install a compiled whl package from Christoph Gohlke's webpage. This whl contains numpy and is linked against mkl. When installing this package, you install both: numpy with the mkl dependencies.

All you have to do is:

  • download the correct whl file (Choose the right Python version and 32/64 file)
  • open a Windows cli with Windows+R and by running inside cmd
  • go to the directory where you have downloaded the whl file, with cd instructions
  • run pip install numpy‑1.XX.Y+mkl‑cp3X‑cp3Xm‑win_amd64.whl

For example, the command can be

pip install numpy‑1.11.3+mkl‑cp35‑cp35m‑win_amd64.whl

You can do it for any package with some code that has to be compiled

Guillaume Jacquenot
  • 11,217
  • 6
  • 43
  • 49
  • 1
    Agree, this was the only way I've got it installed without any "magic" required. – Mikhail Churbanov Dec 19 '16 at 07:50
  • Does anyone know how to change where the MKL libraries are installed? It seems they're going to a dir where I don't have access privileges, and I keep with a basic numpy, no mkl. – Daniel Möller Aug 18 '17 at 13:08
  • 1
    You can use `--user` flag to install it in a folder with your permission rights – Guillaume Jacquenot Aug 18 '17 at 13:14
  • What do you mean --user flag. Where to add this. – Brana Nov 28 '17 at 23:56
  • Anyways your solution doesn't work. I still do not have MKL installed just numpy. Access is denied for something that contains multiarray.py – Brana Nov 28 '17 at 23:57
  • Why do I need to manually download this specific whl when every other module works fine be default? Maybe this should be it's own question, but just find it very inconsistent. – AnnanFay Mar 17 '18 at 12:52
  • @Annan Because the default version is not linked against library mkl that ensures great performance – Guillaume Jacquenot Mar 17 '18 at 14:43
  • @GuillaumeJacquenot Why is there not two versons? Or why is the default not the one with great performance? I notice there's a module called numpy-mkl but it seems to be out of date. ( https://pypi.python.org/pypi/numpy-mkl ) – AnnanFay Mar 19 '18 at 20:51
2

The easiest way is to install an entire Python distribution with lots of included packages, such as numpy and mkl. I would suggest the Anaconda Python distribution, https://www.continuum.io/downloads

jmd_dk
  • 12,125
  • 9
  • 63
  • 94
0

The above answer does great (+1) and brought me on the right track, but to clear things up even more:

  1. You can download the .whl from here
  2. For choosing the right .whl you need to know numpy‑1.11.3+mkl‑cpXX‑cpXXm‑win_amd64.whl where the XX are actually your python version (e.g. 36 for python version 3.6.x)
  3. Do pip install numpy‑1.11.3+mkl‑cp**XX**‑cp**XX**m‑win_amd64.whl in your cmd window and the place where you did download the .whl into
mrk
  • 8,059
  • 3
  • 56
  • 78