initial betas and simulated annealing

[Eldar]

Just a simple question: Is it helpful to provide initial betas for simanneal optimization routine?

[Doherty]

Supplying initial beta's probably won't hurt, but the benefit will not be as great as with the regular optimization routine.

[bacollier]

Eldar,
I did some tinkering with this in a multi-strata model recently, in my tests, never made much difference (even when I set the initial value to the actual estimate I got from a previous simulated annealing run). it makes sense that the benefit was negligible because (if I understand what Gary did) as you start to approach some point on the likelihood surface then you will suddenly makes a random jump out away from there and see it if tracks back that way. Thus, I think the 'initial' location is not really needed unless I misunderstood something on how MARK does simulate annealing. But, that was just my one example so their may be cases where it works better...

Bret

[Eldar]

Thanks, I was thinking that it will not be too helpful too, but did not want to test as it takes so long.