TalkChess.com

lkaufman wrote:Latest Komodo (with contempt to avoid needless exchanges)

on fast six core vs. FM Bolzoni (FIDE 2187, born 1962), Komodo giving knight (b1) odds but always getting White and winning draws. Time limit 45' + 15". The only similar match on record was Rybka (near Rybka 3, on octal machine about 8 years ago) against FM John Meyer, won by Meyer with a perfect 4 out of 4 score. Without making any prediction about the match myself, I'll just make the comment that IF Komodo manages to draw or win the match, I think it would be a remarkable achievement. Masters normally just resign when they lose a piece for nothing, or even sometimes if they get a pawn for it. If Komodo pulls off a drawn or better result, it would mean that chess is a much deeper and more difficult game than is generally believed. So even if I had no involvement with Komodo, I would root for it just for what it would tell us about how far human play is from what is possible.

Laskos wrote:

lkaufman wrote:Latest Komodo (with contempt to avoid needless exchanges)

Isn't contempt in Komodo similar to drawscore? And now draws are good. Some exchanges might be beneficial in achieving the draw. I am not sure how will you achieve a better score to win and to draw using only the drawscore. If you reasonably fix the goals of Komodo in this match, I would bet on Komodo getting equal or more than 3/6.
quote]

Until now contempt was just another name for drawscore (with sign flipped). But now we have changed Komodo to also use the contempt setting to try to avoid piece exchanges, especially queen exchanges. It's probably similar to what Houdini did. We will probably keep the default setting modest, but for handicap matches it should be pretty aggressive. I'm trying to determine now what setting would likely work best in this match. What do you mean by "fix the goals of Komodo"? Komodo will still aim for a draw when possible if it's significantly behind; if it is not much behind it will almost surely not lose in any case.

lkaufman wrote:

Laskos wrote:

lkaufman wrote:Latest Komodo (with contempt to avoid needless exchanges)

Isn't contempt in Komodo similar to drawscore? And now draws are good. Some exchanges might be beneficial in achieving the draw. I am not sure how will you achieve a better score to win and to draw using only the drawscore. If you reasonably fix the goals of Komodo in this match, I would bet on Komodo getting equal or more than 3/6.
quote]

Until now contempt was just another name for drawscore (with sign flipped). But now we have changed Komodo to also use the contempt setting to try to avoid piece exchanges, especially queen exchanges. It's probably similar to what Houdini did. We will probably keep the default setting modest, but for handicap matches it should be pretty aggressive. I'm trying to determine now what setting would likely work best in this match. What do you mean by "fix the goals of Komodo"? Komodo will still aim for a draw when possible if it's significantly behind; if it is not much behind it will almost surely not lose in any case.

For example, a goal would be to get a simpler, surer draw than to complicate things for a possible win when at mild disadvantage. I understood that a draw is equal to a win.

Jesse Gersenson wrote:1 out of 6, one tactical oversight.

Laskos wrote:

lkaufman wrote:

Laskos wrote:

lkaufman wrote:Latest Komodo (with contempt to avoid needless exchanges)

Isn't contempt in Komodo similar to drawscore? And now draws are good. Some exchanges might be beneficial in achieving the draw. I am not sure how will you achieve a better score to win and to draw using only the drawscore. If you reasonably fix the goals of Komodo in this match, I would bet on Komodo getting equal or more than 3/6.
quote]

Until now contempt was just another name for drawscore (with sign flipped). But now we have changed Komodo to also use the contempt setting to try to avoid piece exchanges, especially queen exchanges. It's probably similar to what Houdini did. We will probably keep the default setting modest, but for handicap matches it should be pretty aggressive. I'm trying to determine now what setting would likely work best in this match. What do you mean by "fix the goals of Komodo"? Komodo will still aim for a draw when possible if it's significantly behind; if it is not much behind it will almost surely not lose in any case.

For example, a goal would be to get a simpler, surer draw than to complicate things for a possible win when at mild disadvantage. I understood that a draw is equal to a win.

Once the score is close to zero Komodo won't lose in any case; the problem is how to play when we are still totally lost. Obviously Komodo should avoid most piece trades when down a piece, but it's not so obvious whether to try for open positions or not. Open positions favor the computer, but also favor simplification. What other good general rules apply when down a piece? Not much theory on that, since most GMs just resign when down a piece for nothing.