Sunday: scaling, runs

laptop: ubuntu upgrades: 10.04 LTS -> 10.10 -> 11.04, geesh.

Set up for running primates.R in likelihood mode with abstracted parallelization:

Carver trouble getting wrightscape installed: even after module load gsl, cannot find gsl library. Testing on zero in mpi mode, 16. Testing on farm, mpi mode with snow, 16 cores.

hmm, high (9%) memory usage on primates.R on zero… monitoring situation closely… Only .3% on farm so far, which has 24G instead of zero’s 32Gig RAM..

farm runs at 16 cores, bm vs ouch: successful. Little difference in differing optima between new and old world monkeys vs brownian motion:

  • Running bm vs alphas models on primates, 64 cpu on farm

  • Running sigmas vs alphas model as primates2.R, 64 cpu on farm

estimate of sigmas model on primates data is very slow relative to alpha, etc.  hmm..

  • Running ou vs alphas on parrotfish.R 64 cpu on farm

(looks like sigmas v alphas but note sigmas is defined as all global, i.e. 1-peak ou, in the model.spec).  Add labrid data to git tracking to facilitate this.  Dramatically better log likelihood on the alphas model, easily supported by the bootstrap: (trait is size-corrected protrusion).

plotting for parameter distributions is supported from wrightscape, to handle the regimes grouping on matching plots.  A more basic/general one is found in pmc pots.  Should  add a decent general plotting of parameters to mcmcTools, as well as ROC curve tools.

Updated socialR a bit to allow more minimal details specified in upload, and to work better with reporting from head node following a cluster run.