Intel® Math Kernel Library (MKL) Error using Python

I first encountered this when I was running Auto-ARIMA.

Intel MKL ERROR: Parameter 4 was incorrect on entry to DLASCL.

Let me share my solution with you.

1. Update conda

You need at least conda 4.1.11, so first update your conda.

conda update conda

2. Tell conda to choose Intel packages over default packages, when available.

Type the code below to choose Intel packages over default packages:

conda config --add channels intel

After updating you’ll be prompted: “Warning: ‘intel’ already in ‘channels’ list, moving to the top”

3. Installing the Intel® Distribution for Python

Get Joseph Magiya’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

I recommend you create a new environment when installing. Type the code in your command prompt if you want the full Intel distribution:

conda create -n idp intelpython3_full python=3

This will take some time (have some water, take deep breaths while you imagine all the things that will now be possible).

After that is done — you should see this:

Installing the Intel® Distribution for Python

BONUS

If you’re doing this, go all the way!

Installing the Intel® Performance Libraries

If you want to build a native extension that directly uses the performance libraries, then you will need to obtain a development package that contains header files and static libraries.

Packages available:
mkl — Intel® Math Kernel Library (Intel® MKL) dynamic runtimes

mkl‑devel — Intel® MKL dynamic runtimes and headers for building software

mkl‑static — Intel® MKL static libraries and headers for building software

mkl‑include — Intel® MKL headers only. Automatically installed along with development packages

Make sure the Intel channel is added to your conda configuration (see above). Then install any of our available performance libraries using “conda install” as normal, such as:

conda install mkl-devel

Cheers!