Using single core even when n_jobs=4 is used

[MayukhSobo]

I am seeing

Performing t-SNE using 1 cores.
Using no_dims = 2, perplexity = 30.000000, and theta = 0.500000
Computing input similarities...
Building tree...

this verbose message even when my n_jobs=4

[jeffThompson]

I see the same issue on Mac, installed with pip.

Attempting to run on 8 cores, but says it's only running on 1.

[jeffThompson]

Not a great solution, but in the short term this fork has Mac-specific install instructions that worked for me: https://github.com/sg-s/Multicore-TSNE

[sg-s]

ouch -- i just nuked that repo, thinking this would be OK and now i find this bug. fml

update: repo restored

[jwittenbach]

I had this problem on my Mac as well. After reading the notes form @sg-s (huge help!), I was able to come up with a relatively simple fix.

1. install a more updated version of the C/C++ compilers using homebrew: brew install gcc
2. find the executables -- for me it installed gcc-8 and g++-8 to /usr/local/bin/ (found via which gcc-<version>)
3. set them as your default C and C++ compilers: export CC=/path/to/gcc-8; export CXX=/path/to/g++-8
4. clone the Multicore-TSNE repo and pip install . from the top level

It's possible that you only need to set he correct C++ compiler (and not the C compiler), but I didn't test trying to make it simpler.

[ohld]

Just faced the same problem and the @jwittenbach solution helped! Thanks!

[chm90]

The solution by @jwittenbach also worked for me. The below formulation the command of point 3. might be easier to use export CC=$(which gcc-8); export CXX=$(which g++-8).

[vishalmhjn]

@jwittenbach these settings fail to create the .so file in the installation module.