Sets a plan selector to be used for choosing plans for removal, if they
have more plans than the specified maximum. This defaults to
WorstPlanForRemovalSelector.
Thoughts about using the logit-type selectors with negative logit model scale parameter:
- Look at one agent.
- Assume she has the choice between n different plans.
- (Continuous) fraction f(i) of plan i develops as (master equation)
df(i)/dt = - p(i) * f(i) + 1/n
where p(i) is from the choice model.
- Steady state solution (df/dt=0) f(i) = 1/n * 1/p(i) .
- If p(i) = e-b*U(i), then f(i) = eb*U(i) / n . Or in words:
If you use a logit model with a minus in front of the beta for plans removal, the resulting steady state distribution is
the same logit model with normal beta.
The implication seems to be: divide the user-configured beta by two, use one half for choice and the other for plans removal.
The path size version still needs to be tested (both for choice and for plans removal).