Re: [gnugo-devel] using MPI for gnugo.

[Gunnar Farnebäck]

Darren Cook wrote:
>> code. The only thing that matters in the short term for making GNU Go
>> stronger is to catch up with the Monte Carlo development, especially
>> make the Monte Carlo code scale better to larger boards. Key ideas to
>> look into there are RAVE (rapid action value estimation),
>> progressive widening, and improved move ordering.
>
> And then rename it to be YAMP "Yet Another MonteCarlo Program" ;-).

No, then comes the interesting possibilities of combining it with the
traditional techniques to get more out of both approaches. I believe
for example that Monte Carlo simulations, and possibly rather few of
them, would be a great way to estimate dragon weakness much more
reliably than today. In the other direction local search results
should be quite powerful to guide the Monte Carlo search, although
something like proof number search probably would be better suited
than the reading we're doing today.

> From the point of view of general computer go improvements, in other
> words, someone, anyone, making a really strong computer go program, the
> fact that GnuGo is a traditional style go program has kept the MCTS
> programs honest.

What do you mean?

> Unfortunately traditional style programs don't scale so well. If the
> talented programmers here could put emotional attachments aside, it
> might be best to say GnuGo is now done, and to spend their efforts on
> another open source program like Fuego [1].

There is no such thing as a done computer go program.

/Gunnar