I've been spending most of my discretionary work time over the last couple of weeks grinding through the process of getting my computer models up and running again. The main challenges have involved converting my models from Repast J to Repast Simphony. While the Java components are the same, the "world" of the models is structured differently in Simphony. So I've had to try to figure out how to re-connect the various parts of the models using what Simphony calls "context." I can't yet say I fully understand how "context" works, but I got my ForagerNet3_Demography_V3 (FN3_D_V3) model up and running by trial and error and looking at examples of code from other models.
Since I haven't figured out to configure the model to use the batch run GUI (I haven't even been able to find it yet, although it apparently exists somewhere in Eclipse), I've been using a primitive parameters file to do batch runs. As I wrote on Friday, these batches would throw an "Out of Memory" error and freeze up around the fortieth run. That suggested some kind of memory leak where object produced during a run were not being deleted before the next run. The gradual accumulation of unused objects eats up the memory until there isn't any left to use to run the model, then it dies.
After going through the code several times and trying a bunch of options, I think I finally found the culprit(s) and made the corrections. I set the model to run 500 times a couple of days ago, and things seemed to be chugging along just fine (when I got to my office this morning it was at run number 720-something, so clearly I didn't have the "stop" command implemented correctly). Anyway, I've now got a batch of new data that I can compare with data produced by the model when it was implemented in Repast J. The figure below compares old model data (left) with new model data (right). It is gratifying to see the model is behaving the same.
Since I haven't figured out to configure the model to use the batch run GUI (I haven't even been able to find it yet, although it apparently exists somewhere in Eclipse), I've been using a primitive parameters file to do batch runs. As I wrote on Friday, these batches would throw an "Out of Memory" error and freeze up around the fortieth run. That suggested some kind of memory leak where object produced during a run were not being deleted before the next run. The gradual accumulation of unused objects eats up the memory until there isn't any left to use to run the model, then it dies.
After going through the code several times and trying a bunch of options, I think I finally found the culprit(s) and made the corrections. I set the model to run 500 times a couple of days ago, and things seemed to be chugging along just fine (when I got to my office this morning it was at run number 720-something, so clearly I didn't have the "stop" command implemented correctly). Anyway, I've now got a batch of new data that I can compare with data produced by the model when it was implemented in Repast J. The figure below compares old model data (left) with new model data (right). It is gratifying to see the model is behaving the same.
The data on the left are from a paper of mine ("The Sensitivity of Demographic Characteristics to the Strength of the Population Stabilizing Mechanism in a Model Hunter-Gatherer System") that will be published in an upcoming volume titled Uncertainty and Sensitivity Analysis in Archaeological Computational Modeling (edited by Marieka Brouwer Burg, J. H. M. Peeters, and William A. Lovis; Springer).
I'm happy that this model is back in business. I'll do some more testing to make certain everything is working, then I'll clean up the code and make it available on this website and under my profile at OpenABM. I plan to use this model for some work on the demographic viability of small populations and, perhaps, to push ahead with exploring demography, mortality, and fertility during the Middle Paleolithic.
I'm happy that this model is back in business. I'll do some more testing to make certain everything is working, then I'll clean up the code and make it available on this website and under my profile at OpenABM. I plan to use this model for some work on the demographic viability of small populations and, perhaps, to push ahead with exploring demography, mortality, and fertility during the Middle Paleolithic.
RSS Feed
