[computer-go] Monte Carlo Scalability

[Don Dailey]

I put up another version of GenericMC, a very generic Monte Carlo
reference program.  

This version samples 100,000 games instead of 10,000 and I call it
GenericMC_100K

Since it is doing 10 times as many games, it will be interesting to see
how much this improves the strength.  

Of course the primitive heuristic that it uses does not scale
infinitely, but I do expect a significant increase in strength at these
levels.  

AnchorMan uses the same basic techniques of this primitive Monte Carlo
program,  but is "tricked out" with a few more goodies such as some
trivial (but useful) tactical helpers and the  "all moves as first"
heuristic which makes it play significantly better with fewer games.   

AnchorMan is already well into the diminishing returns curve and will
not play much better even with significantly more thinking time.

I am rather interested to see if GenericMC can do better than AnchorMan
since it doesn't use "all moves as first" which surely hurts it if the
level is high enough.

- Don