1

In Windows, I have a Python 3.6 script that uses the cvxopt package which needs numpy + MKL for it to work. numpy + MKL was successfully installed using the prebuilt binaries here.

Problem: Now the script also needs to be run in Ubuntu 16.04, but pip install numpy installs the version of numpy without MKL.

How can I install numpy+MKL on a Linux system to be able to use the cvxopt package?

Nyxynyx
  • 61,411
  • 155
  • 482
  • 830
  • You can build it from source with linked to MKL – percusse Oct 31 '17 at 20:13
  • Easy solution: use anaconda's python-distribution. (although you did not make clear why you want MKL; cvxopt will work fine with alternatives) – sascha Nov 02 '17 at 15:56
  • @sascha I created my virtualenv using `conda create --name myEnv python=3.6`, maybe this is why it doesnt include numpy+MKL? – Nyxynyx Nov 02 '17 at 20:23
  • @sascha What alternatives are there in order to use `cvxopt`? – Nyxynyx Nov 02 '17 at 20:24
  • It does not matter if your numpy is using / linked against MKL, OpenBlas, Atlas (and even on default Ubuntu this should work already when just using pip install numpy)... Your conda call should be MKL-based. Don't expect it to have a name including MKL. It's the default by now. Look here at SO for reading out numpy's setup. – sascha Nov 02 '17 at 20:48

0 Answers0