Open harmonyrose opened 9 months ago
Tell me if I'm wrong, but I feel like we discussed this earlier in the semester and our reasoning led us to "aw, let's just run it for a fixed number of iterations." Is your motive for doing this one of performance? (Like, sims are taking too long to run?)
Another argument for a fixed number of iterations is it makes the analysis easier. Consider: if you ran two sims, and one had 8 elections before it terminated and the other had 17, it's perhaps difficult to do anything useful with the data from the 9 "extra" elections of the second sim, since there's nothing to directly compare them to. If every sim has the same number of elections it might simplify the analysis.
One motive is performance for when we get to doing param sweeps and batch runs.
I suppose I was thinking about analyzing the results "at simulation's end", once everything has settled, as was the analysis with CI2. As it is now, the simulation isn't always "settling" after 8 elections.
If we are analyzing elections over time instead, I agree it's easier to have a fixed number of iterations and we don't need to worry about when the simulation is "done."
I wonder if we need to simply get you access to a faster machine. (i.e., with more cores.) We used to have a dedicated 64-core server for DS research, but it became obsolete. :( We could spin up a Google Cloud or Amazon EC2 instance, though -- I'll ask Dr. A if we have any money for that.
Number of buckets quiesces? Number of viable candidates quiesces? Fewer than a certain number of inter-agent influences?