Komodo knight odd match vs FM Victor Bolzoni : report

[mjlef]

...Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05".

Are material values static values (think, knight = 3)? Or are they more complicated, like perhaps a function of how many pieces are on the board, or how many pieces have moved from the normal, or defined, starting position, etc?

Komodo modifies piece value based on game stage (related to remaining material), and a whole lot of other things. We try to make Komodo's eval as smooth as possible.

[lkaufman]

...Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05".

Are material values static values (think, knight = 3)? Or are they more complicated, like perhaps a function of how many pieces are on the board, or how many pieces have moved from the normal, or defined, starting position, etc?

They depend on what other pieces and pawns are on the board, but not on the position of those pieces.

[Laskos]

The difference in your results is huge, and I doubt that it would go away entirely at a longer tc. I would suggest removing the White queen knight from the 7 move position, and making some other small change to the position to get close to a zero score from Komodo, and then running an equal time match. I expect that the extra piece side will win the match comfortably. Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05". I'm at a loss to explain why. Anyone have any ideas how to reconcile these seemingly contradictory facts?

It seems that with these "equal", very unbalanced positions, Komodo behaves fairly well, although it's hard to test with my notebook. Taking this position which shows +0.67 for White after 10 minutes of thinking:
[d]rnbqkbnr/pppppppp/8/3P4/2B1PB2/5N2/PPP1QPPP/2KR3R w kq - 0 1

The match outcome at ultra-fast is reasonable, just a bit depressed for White considering its +0.67 advantage:

Games : 100 (finished)

White Wins : 48 (48.0 %)
Black Wins : 36 (36.0 %)
Draws : 16 (16.0 %)
Unfinished : 0

With this position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

The match outcome is again reasonable, and again White is a bit subdued, in 0.00 evaluated position:

Games : 100 (finished)

White Wins : 36 (36.0 %)
Black Wins : 48 (48.0 %)
Draws : 16 (16.0 %)
Unfinished : 0


It seems that the positional overvaluation is in the middle of these 2 positions, about 0.33 or some 10% of positional factors, if these two results mean something (too few games, too short TC). It's hard to interpret what this all combined amounts to, the positional +3.90 advantage alone showed much less importance than a real knight, but now they pretty much compensate each other, not totally, though. Maybe longer time control corrects for these anomalies.

[AdminX]

The difference in your results is huge, and I doubt that it would go away entirely at a longer tc. I would suggest removing the White queen knight from the 7 move position, and making some other small change to the position to get close to a zero score from Komodo, and then running an equal time match. I expect that the extra piece side will win the match comfortably. Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05". I'm at a loss to explain why. Anyone have any ideas how to reconcile these seemingly contradictory facts?

It seems that with these "equal", very unbalanced positions, Komodo behaves fairly well, although it's hard to test with my notebook. Taking this position which shows +0.67 for White after 10 minutes of thinking:
[d]rnbqkbnr/pppppppp/8/3P4/2B1PB2/5N2/PPP1QPPP/2KR3R w kq - 0 1

The match outcome at ultra-fast is reasonable, just a bit depressed for White considering its +0.67 advantage:

Games : 100 (finished)

White Wins : 48 (48.0 %)
Black Wins : 36 (36.0 %)
Draws : 16 (16.0 %)
Unfinished : 0

With this position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

The match outcome is again reasonable, and again White is a bit subdued, in 0.00 evaluated position:

Games : 100 (finished)

White Wins : 36 (36.0 %)
Black Wins : 48 (48.0 %)
Draws : 16 (16.0 %)
Unfinished : 0


It seems that the positional overvaluation is in the middle of these 2 positions, about 0.33 or some 10% of positional factors, if these two results mean something (too few games, too short TC). It's hard to interpret what this all combined amounts to, the positional +3.90 advantage alone showed much less importance than a real knight, but now they pretty much compensate each other, not totally, though. Maybe longer time control corrects for these anomalies.

Can I just tell you Kai,

It's examples like this which is why I have had to unlearn most of what I thought I knew about the game. Tarrasch would die again if he were to return, and Nimzowitsch would be laughing his butt off.

[lkaufman]

The difference in your results is huge, and I doubt that it would go away entirely at a longer tc. I would suggest removing the White queen knight from the 7 move position, and making some other small change to the position to get close to a zero score from Komodo, and then running an equal time match. I expect that the extra piece side will win the match comfortably. Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05". I'm at a loss to explain why. Anyone have any ideas how to reconcile these seemingly contradictory facts?

It seems that with these "equal", very unbalanced positions, Komodo behaves fairly well, although it's hard to test with my notebook. Taking this position which shows +0.67 for White after 10 minutes of thinking:
[d]rnbqkbnr/pppppppp/8/3P4/2B1PB2/5N2/PPP1QPPP/2KR3R w kq - 0 1

The match outcome at ultra-fast is reasonable, just a bit depressed for White considering its +0.67 advantage:

Games : 100 (finished)

White Wins : 48 (48.0 %)
Black Wins : 36 (36.0 %)
Draws : 16 (16.0 %)
Unfinished : 0

With this position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

The match outcome is again reasonable, and again White is a bit subdued, in 0.00 evaluated position:

Games : 100 (finished)

White Wins : 36 (36.0 %)
Black Wins : 48 (48.0 %)
Draws : 16 (16.0 %)
Unfinished : 0


It seems that the positional overvaluation is in the middle of these 2 positions, about 0.33 or some 10% of positional factors, if these two results mean something (too few games, too short TC). It's hard to interpret what this all combined amounts to, the positional +3.90 advantage alone showed much less importance than a real knight, but now they pretty much compensate each other, not totally, though. Maybe longer time control corrects for these anomalies.

Can I just tell you Kai,

It's examples like this which is why I have had to unlearn most of what I thought I knew about the game. Tarrasch would die again if he were to return, and Nimzowitsch would be laughing his butt off.

I'm not sure why you say that. The classical rule is that you need three tempi for a pawn and three pawns for a knight. In these examples White has about 8 tempi for the knight, so he should be a tad worse according to the classical ideas. For an example of what was considered adequate compensation for a knight in the 1800s, look up the Muzio Gambit, which i think is still considered about equal.

[lkaufman]

The difference in your results is huge, and I doubt that it would go away entirely at a longer tc. I would suggest removing the White queen knight from the 7 move position, and making some other small change to the position to get close to a zero score from Komodo, and then running an equal time match. I expect that the extra piece side will win the match comfortably. Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05". I'm at a loss to explain why. Anyone have any ideas how to reconcile these seemingly contradictory facts?

It seems that with these "equal", very unbalanced positions, Komodo behaves fairly well, although it's hard to test with my notebook. Taking this position which shows +0.67 for White after 10 minutes of thinking:
[d]rnbqkbnr/pppppppp/8/3P4/2B1PB2/5N2/PPP1QPPP/2KR3R w kq - 0 1

The match outcome at ultra-fast is reasonable, just a bit depressed for White considering its +0.67 advantage:

Games : 100 (finished)

White Wins : 48 (48.0 %)
Black Wins : 36 (36.0 %)
Draws : 16 (16.0 %)
Unfinished : 0

With this position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

The match outcome is again reasonable, and again White is a bit subdued, in 0.00 evaluated position:

Games : 100 (finished)

White Wins : 36 (36.0 %)
Black Wins : 48 (48.0 %)
Draws : 16 (16.0 %)
Unfinished : 0


It seems that the positional overvaluation is in the middle of these 2 positions, about 0.33 or some 10% of positional factors, if these two results mean something (too few games, too short TC). It's hard to interpret what this all combined amounts to, the positional +3.90 advantage alone showed much less importance than a real knight, but now they pretty much compensate each other, not totally, though. Maybe longer time control corrects for these anomalies.

I think that what's going on is that with a development advantage you need to play sharply to fully exploit it, which favors the stronger player. So even if eight tempi (let's say) are adequate compensation for a knight, an eight move handicap requires more skill to exploit than an extra piece.
It would be very interesting to test your hypothesis that longer tc favors the attacker. If you double or quadruple your time limit we could find out. It's not obvious to me which side benefits more from longer tc, the attacker or the side with more material.

[h1a8]

The difference in your results is huge, and I doubt that it would go away entirely at a longer tc. I would suggest removing the White queen knight from the 7 move position, and making some other small change to the position to get close to a zero score from Komodo, and then running an equal time match. I expect that the extra piece side will win the match comfortably. Yet any attempt to simply raise material values gets a negative elo result here at ultra fast levels like 10" +.05". I'm at a loss to explain why. Anyone have any ideas how to reconcile these seemingly contradictory facts?

It seems that with these "equal", very unbalanced positions, Komodo behaves fairly well, although it's hard to test with my notebook. Taking this position which shows +0.67 for White after 10 minutes of thinking:
[d]rnbqkbnr/pppppppp/8/3P4/2B1PB2/5N2/PPP1QPPP/2KR3R w kq - 0 1

The match outcome at ultra-fast is reasonable, just a bit depressed for White considering its +0.67 advantage:

Games : 100 (finished)

White Wins : 48 (48.0 %)
Black Wins : 36 (36.0 %)
Draws : 16 (16.0 %)
Unfinished : 0

With this position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

The match outcome is again reasonable, and again White is a bit subdued, in 0.00 evaluated position:

Games : 100 (finished)

White Wins : 36 (36.0 %)
Black Wins : 48 (48.0 %)
Draws : 16 (16.0 %)
Unfinished : 0


It seems that the positional overvaluation is in the middle of these 2 positions, about 0.33 or some 10% of positional factors, if these two results mean something (too few games, too short TC). It's hard to interpret what this all combined amounts to, the positional +3.90 advantage alone showed much less importance than a real knight, but now they pretty much compensate each other, not totally, though. Maybe longer time control corrects for these anomalies.

I think that what's going on is that with a development advantage you need to play sharply to fully exploit it, which favors the stronger player. So even if eight tempi (let's say) are adequate compensation for a knight, an eight move handicap requires more skill to exploit than an extra piece.
It would be very interesting to test your hypothesis that longer tc favors the attacker. If you double or quadruple your time limit we could find out. It's not obvious to me which side benefits more from longer tc, the attacker or the side with more material.

That's why you are smart. Time is needed to figure out how to take advantage of tempi advantage. Skill is needed (in the form of time).

[Laskos]

I think that what's going on is that with a development advantage you need to play sharply to fully exploit it, which favors the stronger player. So even if eight tempi (let's say) are adequate compensation for a knight, an eight move handicap requires more skill to exploit than an extra piece.
It would be very interesting to test your hypothesis that longer tc favors the attacker. If you double or quadruple your time limit we could find out. It's not obvious to me which side benefits more from longer tc, the attacker or the side with more material.

Nice, maybe that's why while time handicapped the positional advantage is harder to convert than material. I can confirm, though with difficulty on my notebook, in not very many games and only one position, that the hypothesis of longer time control favoring the attacker is valid. The games are on weak 2 cores of 1.2GHz each. When I will be back home, I may test more thoroughly.

With the previous position showing 0.00 after 10 minutes:
[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N3P1/PPP1QP1P/1K1R3R w kq - 0 1

Increasing time controls (still very fast):

Code: Select all

2.5''+0.025''
Games        :    100 (finished)

White Wins   :     34 (34.0 %)
Black Wins   :     52 (52.0 %)
Draws        :     14 (14.0 %)
Unfinished   :      0



5''+0.05''
Games        :    100 (finished)

White Wins   :     36 (36.0 %)
Black Wins   :     48 (48.0 %)
Draws        :     16 (16.0 %)
Unfinished   :      0



20''+0.2''
Games        :    100 (finished)

White Wins   :     42 (42.0 %)
Black Wins   :     37 (37.0 %)
Draws        :     21 (21.0 %)
Unfinished   :      0

At longer 20''+0.2'' the outcome already surpasses the eval of 0.00. My guess is that to longer TC it may be beneficial to even increase the importance of positional factors in Komodo. This might be a problem of all engines which have to pass first very short time control, it may harm there, but it probably scales well to longer TC.

[AdminX]

I'm not sure why you say that. The classical rule is that you need three tempi for a pawn and three pawns for a knight. In these examples White has about 8 tempi for the knight, so he should be a tad worse according to the classical ideas. For an example of what was considered adequate compensation for a knight in the 1800s, look up the Muzio Gambit, which i think is still considered about equal.

Hi Larry,

Well when you look at the position, white has more than tempi, he also has a great space advantage as well as control of the center and a big lead in development. That's why I say that, according to classical rules white's position should be crushing. At least visually in the past that's what I would have thought.

[Joerg Oster]

I'm not sure why you say that. The classical rule is that you need three tempi for a pawn and three pawns for a knight. In these examples White has about 8 tempi for the knight, so he should be a tad worse according to the classical ideas. For an example of what was considered adequate compensation for a knight in the 1800s, look up the Muzio Gambit, which i think is still considered about equal.

Hi Larry,

Well when you look at the position, white has more than tempi, he also has a great space advantage as well as control of the center and a big lead in development. That's why I say that, according to classical rules white's position should be crushing. At least visually in the past that's what I would have thought.

Me too. OTOH, the starting position is very solid.
I think I would go for Bxf7+!
If you allow Black to consolidate his position, all the mentioned advantages vanish into thin air.