Return Match for Komodo

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
lkaufman
Posts: 3187
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Return Match for Komodo

Post by lkaufman » Wed Jul 27, 2016 1:23 am

About a year ago, Komodo played its first handicap match with a GM, namely Petr Neuman (current FIDE 2446, born 1978), giving two pawns handicap (c2 and f2). Komodo lost badly, getting just two draws and three losses in five games. Fortunately I didn't get discouraged by that result, and since then we have played 52 games giving handicaps to GMs with only a single loss, to Simon Williams at the excessive handicap of f7 pawn and three moves. Twelve of these were at two pawns handicap (various pawns) with the result six wins for Komodo and six draws. So the question arises, how is it that one player, the second-lowest rated GM to play Komodo, won three games without a loss while no other GM could win a single game with two extra pawns?
We have decided to try to answer this question by giving Neuman a rematch, just as proposed in this forum by Kai, with the choice of the two pawns to be the same as in our latest match with Erenburg, namely e2 every game, plus b2,c2,d2, and f2 once each. If he wins or even draws the match, it would suggest that his technique in winning a won game is much better than his rating would imply. If he loses, especially by more than one game, it would indicate that his initial success was due to any or all of the following factors:
1. He learned from each game how to win at that specific handicap, which didn't change.
2. The specific c2,f2 handicap is much larger than other two pawn handicaps.
3. We introduced Contempt after this match.
4. Komodo has improved in the last year.

The match is set for Aug.1 and 2 at 11a.m. Eastern Daylight Time and 2 p.m. EDT each day. It will take place in the Czech Republic with our webmaster Jesse Gersenson operating, connected remotely to my 24 core machine. Time limit is the usual 45' + 15". Broadcast will be from our site with the moves also being shown (without commentary this time) on chess.com. More details about the broadcast closer to match time.
Komodo rules!

pilgrimdan
Posts: 361
Joined: Sat Jul 02, 2011 8:49 pm

Re: Return Match for Komodo

Post by pilgrimdan » Wed Jul 27, 2016 2:15 am

lkaufman wrote:About a year ago, Komodo played its first handicap match with a GM, namely Petr Neuman (current FIDE 2446, born 1978), giving two pawns handicap (c2 and f2). Komodo lost badly, getting just two draws and three losses in five games. Fortunately I didn't get discouraged by that result, and since then we have played 52 games giving handicaps to GMs with only a single loss, to Simon Williams at the excessive handicap of f7 pawn and three moves. Twelve of these were at two pawns handicap (various pawns) with the result six wins for Komodo and six draws. So the question arises, how is it that one player, the second-lowest rated GM to play Komodo, won three games without a loss while no other GM could win a single game with two extra pawns?
We have decided to try to answer this question by giving Neuman a rematch, just as proposed in this forum by Kai, with the choice of the two pawns to be the same as in our latest match with Erenburg, namely e2 every game, plus b2,c2,d2, and f2 once each. If he wins or even draws the match, it would suggest that his technique in winning a won game is much better than his rating would imply. If he loses, especially by more than one game, it would indicate that his initial success was due to any or all of the following factors:
1. He learned from each game how to win at that specific handicap, which didn't change.
2. The specific c2,f2 handicap is much larger than other two pawn handicaps.
3. We introduced Contempt after this match.
4. Komodo has improved in the last year.

The match is set for Aug.1 and 2 at 11a.m. Eastern Daylight Time and 2 p.m. EDT each day. It will take place in the Czech Republic with our webmaster Jesse Gersenson operating, connected remotely to my 24 core machine. Time limit is the usual 45' + 15". Broadcast will be from our site with the moves also being shown (without commentary this time) on chess.com. More details about the broadcast closer to match time.
awesome...

unfortunately... the result may not go too well for the human...

JJJ
Posts: 1260
Joined: Sat Apr 19, 2014 11:47 am

Re: Return Match for Komodo

Post by JJJ » Wed Jul 27, 2016 7:58 am

lkaufman wrote:About a year ago, Komodo played its first handicap match with a GM, namely Petr Neuman (current FIDE 2446, born 1978), giving two pawns handicap (c2 and f2). Komodo lost badly, getting just two draws and three losses in five games. Fortunately I didn't get discouraged by that result, and since then we have played 52 games giving handicaps to GMs with only a single loss, to Simon Williams at the excessive handicap of f7 pawn and three moves. Twelve of these were at two pawns handicap (various pawns) with the result six wins for Komodo and six draws. So the question arises, how is it that one player, the second-lowest rated GM to play Komodo, won three games without a loss while no other GM could win a single game with two extra pawns?
We have decided to try to answer this question by giving Neuman a rematch, just as proposed in this forum by Kai, with the choice of the two pawns to be the same as in our latest match with Erenburg, namely e2 every game, plus b2,c2,d2, and f2 once each. If he wins or even draws the match, it would suggest that his technique in winning a won game is much better than his rating would imply. If he loses, especially by more than one game, it would indicate that his initial success was due to any or all of the following factors:
1. He learned from each game how to win at that specific handicap, which didn't change.
2. The specific c2,f2 handicap is much larger than other two pawn handicaps.
3. We introduced Contempt after this match.
4. Komodo has improved in the last year.

The match is set for Aug.1 and 2 at 11a.m. Eastern Daylight Time and 2 p.m. EDT each day. It will take place in the Czech Republic with our webmaster Jesse Gersenson operating, connected remotely to my 24 core machine. Time limit is the usual 45' + 15". Broadcast will be from our site with the moves also being shown (without commentary this time) on chess.com. More details about the broadcast closer to match time.
Nice. I d say it's a special case. If the human has indeed find a way to win against this kind of handicap by finding a safe path to the end, elo jump and contempt won't change a thing.

So, despite all odds, I want to believe in Petr Neuman and want him to win again. I hope he will train a little to win this match like he did the first time.

I d like to add than contempt and elo jump are the same things against human. If he won against a 3200 elo komodo, he still can against a 3300.

Lyudmil Tsvetkov
Posts: 6036
Joined: Tue Jun 12, 2012 10:41 am

Re: Return Match for Komodo

Post by Lyudmil Tsvetkov » Wed Jul 27, 2016 9:10 am

Nice running all those matches.
Handicap match-ups are always fun.

It should be noted that all single-pawn handicap positions with white on move are simply won for the stronger side, with the exception of possibly white move, missing h2 pawn, as in this case black has some castling problems. White move with missing a2 pawn is easily won for black, even more so other single-pawn handicaps. So the engine is not only handicapped, but starts with a lost position.

The present 4 handicap positions are indeed very funny. My Komodo 8 shows for all of them between 90-120 cps, but 10 to 15 sound developing moves from black should bring the score to at least 200 cps, and another 10 or 15 moves with no black mistake involved to over 400 cps.

So all 4 positions are very easily won for black, objectively. But, bearing in mind that engines very rarely make bigger shallow mistakes, while humans do that often, or very often, I would not be surprised at all if Komodo wins again.

Lyudmil Tsvetkov
Posts: 6036
Joined: Tue Jun 12, 2012 10:41 am

Re: Return Match for Komodo

Post by Lyudmil Tsvetkov » Thu Jul 28, 2016 10:24 am

Indeed, missing f2 and c2 pawns is the worse possible handicap combination at 2 pawns.
Score rises very quickly to 200 and then 400-500cps, but the most unpleasant thing for the side giving the handicap is that it does not have counterplay at all. Black places an outpost in the center, exchanges and wins.

I played some games with this handicap against Komodo 8, and it did not manage to even attack my king side a single time, while with the remaining 4 positions at 2 pawns Komodo frequently attacks me, and sometimes even mates me.

Reason for GM Neumann's success has been the handicap position, and although I naturally wish him success on August 1st, this match will be much much tougher for him.

lkaufman
Posts: 3187
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Return Match for Komodo

Post by lkaufman » Thu Jul 28, 2016 3:00 pm

Lyudmil Tsvetkov wrote:Indeed, missing f2 and c2 pawns is the worse possible handicap combination at 2 pawns.
Score rises very quickly to 200 and then 400-500cps, but the most unpleasant thing for the side giving the handicap is that it does not have counterplay at all. Black places an outpost in the center, exchanges and wins.

I played some games with this handicap against Komodo 8, and it did not manage to even attack my king side a single time, while with the remaining 4 positions at 2 pawns Komodo frequently attacks me, and sometimes even mates me.

Reason for GM Neumann's success has been the handicap position, and although I naturally wish him success on August 1st, this match will be much much tougher for him.
Do you really think c2f2 handicap is worse for White than f2g2? Aside from that one, why should c2f2 be worse than d2f2? I'm trying to determine whether there is something about the c2f2 combination that could be identified as bad, or is it just that f2 plus almost any other pawn is bad?
Komodo rules!

Jesse Gersenson
Posts: 542
Joined: Sat Aug 20, 2011 7:43 am
Contact:

Re: Return Match for Komodo

Post by Jesse Gersenson » Fri Jul 29, 2016 9:41 am

lkaufman wrote:The match is set for Aug.1 and 2 at 11a.m. Eastern Daylight Time and 2 p.m. EDT each day.
Start time is 15:00 and 18:00 Central European time (9am and 12 noon Eastern Daylight time).

Lyudmil Tsvetkov
Posts: 6036
Joined: Tue Jun 12, 2012 10:41 am

Re: Return Match for Komodo

Post by Lyudmil Tsvetkov » Fri Jul 29, 2016 10:00 am

lkaufman wrote:
Lyudmil Tsvetkov wrote:Indeed, missing f2 and c2 pawns is the worse possible handicap combination at 2 pawns.
Score rises very quickly to 200 and then 400-500cps, but the most unpleasant thing for the side giving the handicap is that it does not have counterplay at all. Black places an outpost in the center, exchanges and wins.

I played some games with this handicap against Komodo 8, and it did not manage to even attack my king side a single time, while with the remaining 4 positions at 2 pawns Komodo frequently attacks me, and sometimes even mates me.

Reason for GM Neumann's success has been the handicap position, and although I naturally wish him success on August 1st, this match will be much much tougher for him.
Do you really think c2f2 handicap is worse for White than f2g2? Aside from that one, why should c2f2 be worse than d2f2? I'm trying to determine whether there is something about the c2f2 combination that could be identified as bad, or is it just that f2 plus almost any other pawn is bad?
It is not about the f2 pawn, but the particular c2-f2 combination.

I had not been trying f2-g2 up until now, pushed some tries, as you well know, Komodo starts with 200cps edge, biggest of all, and indeed this is objectively the worse of the 4 upcoming match starting positions, as well as c2-f2.

But f2-g2 is a bit tricky to play, with king safety involved all the time, so for a human should be almost equal to 4 main positions of the match.

f2-c2 is worse of all for the engine, simply because white does not have the whiff of a counterplay, in sharp distincion to all other positions.
It is true that center is blockaded even with d2-f2 pawns, but with d2-f2 white has the additional option of castling long; main reason for lack of counterplay is that with d2-f2 white has chances at least along 1 central open file - e file - while with c2-f2 both center files - d and e - are forcefully blockaded, so white does not even have minimal counterplay.

In any case, this is my experience, other players with different styles migth have different assessment.

Uri Blass
Posts: 8220
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Return Match for Komodo

Post by Uri Blass » Sat Jul 30, 2016 7:32 am

I think that we have not enough opponents for komodo with 2 pawns handicap.

A win for Petr Neuman may also mean that the technique of Erenburg in winning a won game is much worse than his rating would imply.

User avatar
Laskos
Posts: 8354
Joined: Wed Jul 26, 2006 8:21 pm

Re: Return Match for Komodo

Post by Laskos » Sat Jul 30, 2016 12:08 pm

lkaufman wrote:
Lyudmil Tsvetkov wrote:Indeed, missing f2 and c2 pawns is the worse possible handicap combination at 2 pawns.
Score rises very quickly to 200 and then 400-500cps, but the most unpleasant thing for the side giving the handicap is that it does not have counterplay at all. Black places an outpost in the center, exchanges and wins.

I played some games with this handicap against Komodo 8, and it did not manage to even attack my king side a single time, while with the remaining 4 positions at 2 pawns Komodo frequently attacks me, and sometimes even mates me.

Reason for GM Neumann's success has been the handicap position, and although I naturally wish him success on August 1st, this match will be much much tougher for him.
Do you really think c2f2 handicap is worse for White than f2g2? Aside from that one, why should c2f2 be worse than d2f2? I'm trying to determine whether there is something about the c2f2 combination that could be identified as bad, or is it just that f2 plus almost any other pawn is bad?
f2g2 is shown correctly by Komodo as by far the worse in both eval and self-play.

But:
e2f2 shows an eval of -1.64 to depth 34 and -610 ELO points in self-play at blitz time control.
c2f2 shows an eval of -1.61 to depth 34 and -700 ELO points.

Is Komodo eval a bit off here?

Post Reply