Closed jscamac closed 9 years ago
Great idea. I'll start this off as the test case, once you've confirmed we're good to go.
Models all worked. But output size is just to memory intensive and pretty much is unusable with R's post processing functions (maxes out memory when post processing rho combinations). #72.
Have restructured model so that it does not store individual observation log likelihoods, rather the model just collects the summed log likelihoods. Preliminary runs seem to indicate that file size should decrease by about 72% from ~150mb to 40mb.
I could further reduce this if we don't monitor individual species effects (rather we just monitor the hyper parameters of these effects). I think this is reasonable as we are unlikely to examine individual species effects and thus don't care about these parameter estimates.
Thoughts @richfitz @dfalster?
Rerunning test with changes made based on above
Both growth comparison and rho combination models sampled. Everything looking good so far!
Run all models with 10 iterations to ensure all will sample