2000-05 Suicide and stabilisation


Early stabilisation runs (stats0.csv, etc.)

All these experiments were just individual runs (no run sets and no sets of run sets). Each run produced a single results file (stats0.csv for example). I will take the name of the results file as the name of the experiment for now.

I don't have very much documentation on these. There are no log files (of the configuration files), and I don't have any place where I describe what I was trying to do, or what parameters I was using, or anything along those lines at all. Nor do I have any record of the code for about another 5 months (up to about October).

Nevertheless, it appears as though I was trying to get the simulation to produce stable runs without suicide here. I was also already aware of the bodylength issue (although, I think I remember being aware of that issue from the LISP days). That is, there was a bias in the crossover that would lead to longer and longer bodylengths.

stats118: Started including suicide. I think this was after I had gotten the simulation stable and reasonably bug-free.

Most runs were stable, only a handful died out (used getvalue.py to find this out).

Things start to look quite solid by stats442. By about this time, I had also included a genetic diversity measure. (I think this was the one where I compare what randomly selected pairs of agents would do under the same list of conditions. This is a measure of genetic diversity, simply because agents have no interesting devlopment (no interesting ontogenesis).)