search

dattalab / pyhsmm-library-models

library models built on top of pyhsmm
0 stars 1 forks source link

Parallel sampling fails if likelihoods have been precomputed #29

Closed alexbw closed 11 years ago

alexbw commented 11 years ago

If you run a parallel model once, it will precompute the likelihoods. If you run it again, the workers don't seem to take any memory. I'm just judging from looking at the processes in htop. The first run, they each grabbed ~3GB, now they're sitting at 30MB, and the CPU usage is 0 for all.

alexbw commented 11 years ago

This could be a ghost. Lots of weird stuff going on these computers, need a definite test case before making too much of this.

alexbw commented 11 years ago

Can't recapitulate for now. Closing.