Petr Baudis wrote:
> Another factor is time per game. Some people advise for very fast
> games and/or limitation by playout. I do not think either is good idea.
> Your idea is no good if the program beats gnugo in 5% more cases, but
> takes 20% longer per move.

If your program gets slower and you use limitation by playout, your
results can still be relied on if they tell you that your latest idea
_doesn't_ help (which I fear is the common case). While if you use
limitation by time and your program loses more games, it might be the
case that the change would be worthwhile if you could make it run
faster.

So for changes which are expected to make the program slower, perhaps a
sensible approach is:
 - write code with little effort spent on optimisation
 - test with limitation by playout
 - if play doesn't improve, try another idea
 - otherwise, put some effort into making the code fast
 - test with limitation by time.

There's also the issue that with time limits you have to worry about
things like uniform hardware and other processes interfering.


> Currently, I'm using 10:00 or 8:20 SD for 19x19 testing, giving Pachi
> no komi as white. In the past, I was using something like 4:30 SD, but I
> have found that I'm getting quite different picture in many cases when
> testing with longer time limits. Yes, you need to wait much longer
> and/or use much more hardware (I'm making use of idling department
> workstations at nights) to get statistically significant results, but
> that's one of the burdens of a Go programmer. :-)

One possibility to get more games while still using slower play settings
is to prepare a selection of mid-game positions and start some of the
games from there.

Another trick that might save significant time without losing too much
accuracy is to make the program unusually willing to resign when it
thinks it's doing badly, and maybe even to allow it to 'claim a win' if
it's confident it's won. (Though if the time gnugo is taking isn't very
significant, I suppose it would make more sense to just configure the
program to play the endgame quickly.)

-M-
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to