My own thinking about this is very like yours, or at least it was until
I saw it written.
After reading your post it occurred to me that I have been assuming the
algorithm always knows the best move (or at least that it always picks a
move at least as strong as its opponent). For the case of an even game*,
I suppose that in some board positions the algorithm will chose a better
move, in others the opponent will. In the case where the algorithm is
strong, it will get ahead. Either the opponent will choose an inferior
move and weaken its position or it will choose the expected move and
will feel the effect of pondering. However in the case where the
algorithm is weak, its prediction of the opponent's move will be
incorrect (assuming the opponent isn't similarly weak in this
situation). The opponent will then punish the algorithm by playing a
move which is both stronger and not pondered by our algorithm. I can
start to understand how pondering only the one move may lead to
sub-optimal play.
Noting the general nature of your argument, I think the above applies
without loss of generality?
Thank you for helping with my thinking. Odd how somebody stating a
viewpoint similar to mine changes my mind more easily than somebody
stating an opposite view! I'd be interested in any thoughts you have on
this.
Raffles
*here "even" refers to skill level of opponents assuming no pondering,
i.e. without pondering P(black win) = P(white win)
[email protected] wrote:
My thinking about speculative pondering is as follows (I did not read
the previous discussion carefully so perhaps I just repeat an old
argument):
When the programs select the best move to play it also knows what move
is the *most likely best move* for the opponent given search to that
point.
With speculative pondering all pondering goes into searching the reply
to that *most likely best* move.
If the opponent plays a different move, this move will on average be a
worse move than the expected one. Although there was no gain in
pondering, my program has a better than expected position (on average).
In other words: a strong opponent will cause a lot of ponder hits and
speculative pondering is the best way to search effectively. A weak
program will make many mistakes and the method of pondering does not
matter much. I think there is something general to this argument so it
would apply to any kind of game. Pondering on all possible opponent
moves will always spend search on inferior moves.
-Magnus
Quoting Brian Sheppard <[email protected]>:
It's then the algorithm that is responsible, and absolutely not the
specifics of a game.
Exactly, it is the algorithm. I think we agree on all points.
It just happens that the most successful search algorithm for chess
makes speculative pondering work, whereas in Go it is the opposite.
Brian
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
