DiViMe Parallelize new Yunitator

The current Yunitator is relatively slow, it takes a long time to load a model, and then it takes linear time to process an audio file, using a single CPU only. A future version will hopefully require less memory (or a constant amount of memory), so it may be possible to parallelize the evaluation, probably even within python, to speed up wall-clock time on more powerful machines. Requires fix to #44

Oct 09 '18 14:10 fmetze

I can't decide whether this goes in the "LENA match up" (=primary, urgent) or "improvement" project category... My guess the latter?

Oct 09 '18 17:10 alecristia

My guess as well. First make things work, then make them more efficient.

Oct 09 '18 17:10 riebling

Yes, no students on the critical path.

On Oct 9, 2018, at 1:42 PM, Eric Riebling [email protected] wrote:

My guess as well. First make things work, then make them more efficient.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/srvk/DiViMe/issues/49#issuecomment-428284591, or mute the thread https://github.com/notifications/unsubscribe-auth/AEnA8UOswd3zkw32HIdh7nKCr3W3PTYrks5ujOAWgaJpZM4XTefU.

Oct 09 '18 18:10 fmetze

Jaden, I added you on this issue because it bears on one of the 3 related tools, but my reading is that this is not urgent today.

Oct 30 '18 14:10 alecristia