search

divilian / specstar

Combines SPECscape and SPECnet into one project
1 stars 1 forks source link

Determine rules for exogenous "starvation" shock #44

Closed divilian closed 5 years ago

divilian commented 5 years ago

Figure out and articulate exactly how the simulation should behave to investigate the "how robust are agents & protos to starvation events?" question.

divilian commented 5 years ago

Question, @venkatachalapathy: how exactly do we want the simulation to implement the "starvation shock?"

One idea is: add two new simulation parameters, :start_starv_iter and :end_starv_iter. If (say) the former is 50 and the latter is 60, then during iterations 50 through 60, inclusive, no agent will receive their salary. They will still have their metabolic rate subtracted, as usual, and will still have their white noise up-or-down, as usual.

Does this sound reasonable for phase 1?

venkatachalapathy commented 5 years ago

My idea is to first code up the stopping rule; then implement a starvation rule like the one you suggested. The stopping rule is a separate issue #41 that I closed because we have a constructive definition for it. So, in #45, we have to algorithmize both the stopping conditions and starvation rules.

Both are doable but is this what we want to be spend our time on between now and our end of the month deadline, @WheezePuppet and @will-nehrboss ?

divilian commented 5 years ago

Ah, that makes sense.

And yes I do think we want to prioritize this for phase 1, in part because I don't actually think it will take that long to code, once we figure out exactly what to code.

divilian commented 5 years ago

As of yesterday's Zoom call, we have decided on the following for phase 1.

Every simulation can be divided into three stages: S1. The iterations before any agents create a proto. S2. The iterations during which agents create and join protos. S3. The iterations after which all agents are either (a) graph isolates, (b) dead, or (c) in a proto.

The "stopping condition" is when S2 transitions to S3.