Performance difference to the old version

[kahilah]

Hey!

Does anyone have idea why the lates version (master) is slower than the older version available in
https://github.com/danielfrg/tsne/tree/master/tsne/bh_sne_src

I profiled the issue with the mnist2500 data set and the problem lies in the recursive function computeNonEdgeForces.

Is there something fundamentally wrong / bugs in the old version used by that other repository?

[kahilah]

To answer myself: these issues have been briefly/implicitly discussed in issue #18 and in addition performance issues in sptree.cpp have been tackled in fork https://github.com/resero-labs/tsne