I think this is an interesting idea, but the basic premise seems
questionable to me.
The equity values used by bots for given positions are derived from bot
v bot 'perfect' play. Accurate cube decisions require accurate
knowledge of absolute equities. But in real life, human v human games,
you can't know the true equities for various reasons. Will opponent
handle a recube effectively? Will opponent (or player) move the checkers
well in the ensuing positions? Will opponent make correct choices
between going for gammons or playing to save gammons? Will opponent
incorrectly take or drop cubes now or at a later time?
Trying to subjectively modify the bot's grading scale risks losing the
integrity of the scale completely.
Making a play that sacrifices equity, then crediting yourself if it
'works' is playing the results, which we all have learned is not the
right way to think and improve. It can certainly lead to 'Fancy Play
Syndrome' where we outplay ourselves by overdoing it.
GNU and Snowie already tell you the parameters for which it is
appropriate to deviate from the 'proper' cube action ( ie - NoDouble,
but correct to double if opponent will pass 25% of the time).
For match play, are you using the right MET? What is the true gammon
rate for matches between you and your opponents? For example I play a
gentleman in our weekly club tournaments who routinely drops takeable
cubes, doubles and redoubles late, and does not aggressively pursue
gammons when there are slight risks of losing the game. His checkerplay
is basically strong, better than mine in some ways, worse in others.
What MET should we use, and how should I grade my cube decisions? I'm
not really sure because of how all the variables interact (it's tempting
to revert to an older MET with a lower gammon rate), so I just keep my
modern MET in effect.
I think the 'player adjusted' ratings Albert is looking for already
exist in the scoresheet.
That said, I DO feel there should be a rating adjustment for ggraccoon's
insane luck factor!
Albert Silver wrote:
I had thought of 'filtering' this via a pre-evaluationprocess, where GNU
would use my proposal to grade the respective players, and if it saw that
Player 1 would be rated at least Advanced or Expert in cube play then it
would presume the errors were deliberate (without this, it would obviously
say the player was much worse since it would be giving full value to the
deliberate errors). And if the end result was still a very poor grade, then
it would presume the mutual errors were due to lack of skill.
_______________________________________________
Bug-gnubg mailing list
Bug-gnubg@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gnubg
_______________________________________________
Bug-gnubg mailing list
Bug-gnubg@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gnubg
