[computer-go] 19x19 MC improvement

[Adrian Grajdeanu]

> By the way, does anybody know of any nifty tools or heuristics for
> efficient probabilistic multi-parameter optimization? In other words,
> like multi-dimensional optimization, except instead of your function
> returning a deterministic value, it returns the result of a Bernoulli
> trial, and the heuristic uses those trial results to converge as
> rapidly as possible to parameter values that roughly maximize the
> success probability.

I recommend evolutionary algorithms because they are robust on noise and 
don't require a quadratic or linear model for the function they 
optimize. I would go as simple as a ES(1+1) algorithm (a glorified name 
for a simple hill climber that probes randomly for its next step). I 
would also use restarts: run it once until no more improvement is 
apparent, then run it again and again (restarts) a few times (5-10) and 
take the overall optimum found. You'd be surprised how far you can get 
with this method!

Adrian