random seeds (was Re: [gnugo-devel] predictability vs. strength)

[Evan Daniel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Douglas Ridgway wrote:
> There's a tradeoff between predictability and strength. As a way of
> exploring this, in a quick and dirty way, I replaced the 0.01 constant
> random multiplier in value_moves.c with various other values and looked at
> sets of 100 9x9 games against the unmodified code (5.5 komi, alternating
> colors, 3.5.6+, level 10). Of interest is the win percentage and the
> number of unique games.
>
> Eps   p_win/%                 unique
> 0.01  50 (by definition)      88
> 1.0   45                      93
> 10.0  34 (*)                  99

How did you run the games?  I tried repeating your experiment with
twogtp.pike as the controller, and realized that for some reason the
random seeds of the two engines are always the same!  Presumably, this
significantly reduces the randomness in the games, and the number of
unique games.

Actually, now that I look at the code in twogtp.pike, I'm quite unclear
how it gets the new random seed for the new game.  It would seem that it
expects gnugo to decide on a new random seed as a result of either the
boardsize or clear_board command, which twogtp.pike then gets with
get_random_seed.  However, this seems to depend on behavior of GNU Go
that is not specified by the GTP v2 specification.

Can anyone confirm my understanding of the code here, or am I confused
about something?

Evan Daniel
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFAplym0OcbTJFOafIRAssrAJ90HQW6JQdsNq6XNs+3+Pq6f60GkgCdFvmh
mogus456/4Co73dXHoQ/Nvk=
=6UKs
-----END PGP SIGNATURE-----