Re: [gnugo-devel] Patch: improve separation of similarly-valued moves

[Arend Bayer]

Paul wrote:

> I wrote:
> > Arend Bayer wrote:
> > > I actually haven't thought about the problem how hard it would be to 
> > > compute the exact move probabilities in our current scheme, but it
> > > should be doable. If nothing else, a Monte Carlo-approach would work
> > > well enough (i.e. simulating the decision for 1000 gg_rand()s).
> > 
> > It must be quite simple, as long as we can tweak a new mode into
> > engine in which it would _not_ apply random shifts to move values,
> > but instead would report the scales of those shifts to the caller.
> > I haven't looked at the exact math involved, but I'm sure it can
> > be done strictly with theory of probability, not involving any
> > statistical methods of probability assessment.
> 
> Here we are.
> 
> - new function compute_move_probabilities() in `value_moves.c'
> - new GTP command `move_probabilities'

Sorry if I start sounding repetitive, but please add comments.

Only problem I see with your code (apart from redistribution mentioned
stuff) is that it doesn't see the randomness of the fuseki move
generator.

> It would be nice if anyone made use of this in a test suite.  You
> will likely have to implement more GTP commands (or change the one
> I added).

Well as I said, I would think a script to interpret these results
throughout a game is enough IMHO.

Arend