set bounds (paramx2) narrow bounds via data, or like (or in addition to) utilizing a gaussian model to sample the weights of the model and then updating the distribution based on data or perf?

narrow the bounds over time?

eval multiple models (samples of gaussian) to narrow? because then you store the bounds (param x2 or    from 0-pmax with  rescaling) (or gaussian somehow, tightening? std lower, change center?) <- this note feels forced, but elegantly translated enough that i cannot dismiss it outright

gaussian.

sample many points, find which ones do best, reweight, find new std and avg

> reweight

based on perf

you'll have to tighten the distribution HARD at the start right? with a little bit of leighway for expansion

iterative model collapse, collapsing upon a good model, narrowing the std until it becomes more than likely just one value

you could also start from a trained model to test this out even though the idea is that this will train entire models quicker

you would have to incorporate information about each param's effect on the effectiveness of other parameters possibly (not entirely sure)