Create just one ensemble with the biggest practical amount of neurons (according to Terry). Benchmark running the ensemble as is.
Try different ways of splitting the ensemble processing into threads. Benchmark running this ensemble when split up.
Goal: find out if splitting improves performance in any way.
Whoever tackles this can take what we built in src/distribute-proto and try to break up the C ensemble from with nef (ie matrix_multiplication shouldn't change)
Create just one ensemble with the biggest practical amount of neurons (according to Terry). Benchmark running the ensemble as is. Try different ways of splitting the ensemble processing into threads. Benchmark running this ensemble when split up.
Goal: find out if splitting improves performance in any way.