Next Komodo vs human match

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
lkaufman
Posts: 3722
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Next Komodo vs human match

Post by lkaufman » Wed Sep 11, 2019 12:52 am

This was the closest game. Gilden reached a much superior endgame, but missed a tactic.

Komodo rules!

lkaufman
Posts: 3722
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Next Komodo vs human match

Post by lkaufman » Wed Sep 11, 2019 1:03 am

Marcus9 wrote:
Tue Sep 10, 2019 10:03 pm
lkaufman wrote:
Tue Sep 10, 2019 9:51 pm
Yet another crushing win for Komodo. Seems that Contempt and MCTS go very well together! We'll play another game beyond the scheduled four, I want to give him a chance to score.
Fantastic result, clearly ready for the next level!

What is the contempt value for this match?
Even with Komodo MCTS is recommended use half disadvantage points? (About 175 for knight odd)
Yes, I recommend half. Komodo MCTS eval for knight odds is close to -300 after some time (normal Komodo about -350), so I used 150. I could vary it to produce more variety, but there was no need for that in this match. Curiously, at b1 handicap it chose 1.e4 twice but 1.Nf3 once, even though nothing was changed! Well, MP search and MCTS are not deterministic, they don't always play the same. I think I could win a match from Komodo under these same conditions, but that's because I know so much about both handicap play and computer play, not so much due to my present playing strength. Next opponent for knight odds should probably be an IM, perhaps my son or perhaps IM Eugene Meyer or someone else.
Komodo rules!

mjlef
Posts: 1427
Joined: Thu Mar 30, 2006 12:08 pm
Contact:

Re: Next Komodo vs human match

Post by mjlef » Wed Sep 11, 2019 1:15 am

lkaufman wrote:
Wed Sep 11, 2019 1:03 am
Marcus9 wrote:
Tue Sep 10, 2019 10:03 pm
lkaufman wrote:
Tue Sep 10, 2019 9:51 pm
Yet another crushing win for Komodo. Seems that Contempt and MCTS go very well together! We'll play another game beyond the scheduled four, I want to give him a chance to score.
Fantastic result, clearly ready for the next level!

What is the contempt value for this match?
Even with Komodo MCTS is recommended use half disadvantage points? (About 175 for knight odd)
Yes, I recommend half. Komodo MCTS eval for knight odds is close to -300 after some time (normal Komodo about -350), so I used 150. I could vary it to produce more variety, but there was no need for that in this match. Curiously, at b1 handicap it chose 1.e4 twice but 1.Nf3 once, even though nothing was changed! Well, MP search and MCTS are not deterministic, they don't always play the same. I think I could win a match from Komodo under these same conditions, but that's because I know so much about both handicap play and computer play, not so much due to my present playing strength. Next opponent for knight odds should probably be an IM, perhaps my son or perhaps IM Eugene Meyer or someone else.
I just wanted to mention to Komodo owners that the version used in this match had a code change to allow Contempt to work with the MCTS mode. In released Komodos, in MCTS search mode, Contempt is internally set to 0. We did this because we figured that the way MCTS works, it acts a lot like some Contempt is being used. But these matches required higher Contempt values. We did some testing of Komodo's default Contempt in MCTS mode and found it only effects strength by a couple of elo. So we will likely have Contempt usable in a future release, just in case you want to have some big odds games that might require a larger Contempt.

Mark

duncan
Posts: 10313
Joined: Mon Jul 07, 2008 8:50 pm

Re: Next Komodo vs human match

Post by duncan » Wed Sep 11, 2019 7:09 pm

lkaufman wrote:
Tue Sep 10, 2019 11:50 pm
Sixth game was Gilden's best game, but ultimately yet another loss. He reached a normal opening position in the Caro-Kann with White just missing a knight, and it was difficult to imagine how Komodo would even make a fight of it. Komodo managed to trade queen for two rooks, but was still quite lost, but a surprise tactic won the extra knight and ultimately the game. So a perfect six to zero sweep. Clearly, we need a much stronger player or a longer time limit for knight odds to be competitive. It does seem that Komodo MCTS with Contempt (I set it to 150 for this match) is amazingly strong at giving such handicaps. I think we're still a ways off from giving knight odds at this TC to a 2500 GM, but it doesn't look so farfetched any more.
Congratulations on this excellent result. What was Gilden's take on this. Does he need more time, a bigger handicap,lose 20 years?

User avatar
Nordlandia
Posts: 2443
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Next Komodo vs human match

Post by Nordlandia » Wed Sep 11, 2019 7:57 pm

I don't know if it's feasible but many will enjoy a revamp if Chessmaster face some strong GMs without odds. Or even using some of the grandmaster personalities available from selection. Chessmaster can give time odds i suppose.

lkaufman
Posts: 3722
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Next Komodo vs human match

Post by lkaufman » Wed Sep 11, 2019 8:19 pm

duncan wrote:
Wed Sep 11, 2019 7:09 pm
lkaufman wrote:
Tue Sep 10, 2019 11:50 pm
Sixth game was Gilden's best game, but ultimately yet another loss. He reached a normal opening position in the Caro-Kann with White just missing a knight, and it was difficult to imagine how Komodo would even make a fight of it. Komodo managed to trade queen for two rooks, but was still quite lost, but a surprise tactic won the extra knight and ultimately the game. So a perfect six to zero sweep. Clearly, we need a much stronger player or a longer time limit for knight odds to be competitive. It does seem that Komodo MCTS with Contempt (I set it to 150 for this match) is amazingly strong at giving such handicaps. I think we're still a ways off from giving knight odds at this TC to a 2500 GM, but it doesn't look so farfetched any more.
Congratulations on this excellent result. What was Gilden's take on this. Does he need more time, a bigger handicap,lose 20 years?
He requested a rematch at 30' + 10" instead of 15' + 10". He thinks that with the experience of this match plus the longer time control he can be competitive, and I don't disagree. Since 15' + 10" is the time limit of the FIDE World Rapid championship, I think it should be our standard for matches with IMs and GMs, but for players below that level it's not enough even with knight odds, especially for players over age 70. But he didn't make obvious blunders or miss trivial wins (he did miss one chance to force a draw right before losing one game).
Komodo rules!

duncan
Posts: 10313
Joined: Mon Jul 07, 2008 8:50 pm

Re: Next Komodo vs human match

Post by duncan » Wed Sep 11, 2019 11:25 pm

lkaufman wrote:
Wed Sep 11, 2019 8:19 pm
duncan wrote:
Wed Sep 11, 2019 7:09 pm
lkaufman wrote:
Tue Sep 10, 2019 11:50 pm
Sixth game was Gilden's best game, but ultimately yet another loss. He reached a normal opening position in the Caro-Kann with White just missing a knight, and it was difficult to imagine how Komodo would even make a fight of it. Komodo managed to trade queen for two rooks, but was still quite lost, but a surprise tactic won the extra knight and ultimately the game. So a perfect six to zero sweep. Clearly, we need a much stronger player or a longer time limit for knight odds to be competitive. It does seem that Komodo MCTS with Contempt (I set it to 150 for this match) is amazingly strong at giving such handicaps. I think we're still a ways off from giving knight odds at this TC to a 2500 GM, but it doesn't look so farfetched any more.
Congratulations on this excellent result. What was Gilden's take on this. Does he need more time, a bigger handicap,lose 20 years?
He requested a rematch at 30' + 10" instead of 15' + 10". He thinks that with the experience of this match plus the longer time control he can be competitive, and I don't disagree. Since 15' + 10" is the time limit of the FIDE World Rapid championship, I think it should be our standard for matches with IMs and GMs, but for players below that level it's not enough even with knight odds, especially for players over age 70. But he didn't make obvious blunders or miss trivial wins (he did miss one chance to force a draw right before losing one game).
It would be nice if you could offer him 30' + 10" both for him and the computer chess world.

lkaufman
Posts: 3722
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Next Komodo vs human match

Post by lkaufman » Thu Sep 12, 2019 1:36 am

duncan wrote:
Wed Sep 11, 2019 11:25 pm
lkaufman wrote:
Wed Sep 11, 2019 8:19 pm
duncan wrote:
Wed Sep 11, 2019 7:09 pm
lkaufman wrote:
Tue Sep 10, 2019 11:50 pm
Sixth game was Gilden's best game, but ultimately yet another loss. He reached a normal opening position in the Caro-Kann with White just missing a knight, and it was difficult to imagine how Komodo would even make a fight of it. Komodo managed to trade queen for two rooks, but was still quite lost, but a surprise tactic won the extra knight and ultimately the game. So a perfect six to zero sweep. Clearly, we need a much stronger player or a longer time limit for knight odds to be competitive. It does seem that Komodo MCTS with Contempt (I set it to 150 for this match) is amazingly strong at giving such handicaps. I think we're still a ways off from giving knight odds at this TC to a 2500 GM, but it doesn't look so farfetched any more.
Congratulations on this excellent result. What was Gilden's take on this. Does he need more time, a bigger handicap,lose 20 years?
He requested a rematch at 30' + 10" instead of 15' + 10". He thinks that with the experience of this match plus the longer time control he can be competitive, and I don't disagree. Since 15' + 10" is the time limit of the FIDE World Rapid championship, I think it should be our standard for matches with IMs and GMs, but for players below that level it's not enough even with knight odds, especially for players over age 70. But he didn't make obvious blunders or miss trivial wins (he did miss one chance to force a draw right before losing one game).
It would be nice if you could offer him 30' + 10" both for him and the computer chess world.
I expect we'll do so before too long, but I want to get in a 15' + 10" knight odds match with an IM first if possible, and perhaps one or two other opponents for variety.
Komodo rules!

Post Reply