Post date: Sep 04, 2013 5:39:1 PM
Once again the jobs were entering uninterruptable sleep. I killed all of the jobs. John suggested that I write to local space ($TMPDIR) on the compute nodes. There should be enough space and navier and lycaeides, and navier allows long queue jobs. So, I started the runs again with the 10k test on lycaeides (batch queue, 96hrs), and the full jobs on navier (long queue, 448hrs). I also modified the bgsr code slightly to speed things up. I am no longer printing dprep, and I only print selection coefficients and betas if the model allows non-zero parameter values. Finally, I am now using the distances Patrik sent instead of elevation as my covariate (these were correlated at ~ -0.98). The new jobs are all running (30 total) and have ids 8212-8241).