2000-06 Suicide


Early runs continued

stats1142: Somewhere near this run, I turned off suicide again. I don't know why. But it reappears shortly afterwards. After that, it appears I tried to produce statistics on suicide that are grouped by age and health. (And the debugging that went with that.)

stats1157 (rearranged file in stats1157.xls):

- Cyclical food

- Suicide

Had got the statistics grouped by age/health right by this run. Agent population and rate of suicides per epoch looks like the following:

[IMAGE][IMAGE]

The relative proportions of actions (i.e. the action rates) look like this:

[IMAGE][IMAGE]

I've got no idea why agents choose to turn so often. I think agents choose to mate so little because it is non-consensual at this point in my runs. I've also got no idea why suicide seems such a popular action (exceeding 0.1 probability at certain points). (Gee, that's helpful . . .)

A breakdown of suicide by age and health:

[IMAGE][IMAGE]

We can see that agents with lowest health do most of the suicides, and the youngest agents are also the ones that do the most suicides. (These are absolute numbers --- they're not relative to the number of younger agents, older agents, etc.)

Body length and genetic diversity:

[IMAGE][IMAGE]

Body length was quite low on average, making it difficult to explain the high levels of conditionalisation in the above graphs (though I've got no variance figures). Genetic diversity is about what it was in earlier runs (somewhat < 0.01).

I was also exploring fitness in this run. I checked the average fitness, and it was 5.2. That can't be right, because it is not equal to either the average number of children in a stable simulation (2), or the average number of grandchildren (4).