Komodo knight odd match vs FM Victor Bolzoni : report

[Laskos]

[Neither does Komodo, it gives less than 2 important pawns and a bit more than f7 pawn. However, maximum achievement in 8 moves (not beyond 4th rank) is almost twice as large, somewhere at 3 pawns, between 2 important pawns and a knight. This 8-mover position
[D]rnbqkbnr/pppppppp/8/8/2PPPP2/2NB1N2/PP4PP/R1BQ1RK1 w kq - 0 1[/D] is evaluated by Komodo at +3.15 on my lowly netbook after 6 minutes, with e5 best move. Without any handicap games testing, it's apparently some 1000 ELO points handicap. The rule of thumb seems to be a bit more than 100 ELO points for each available additional move of his choice to GM. I don't see how to prepare Komodo against repeated opening "scheme" of GM if the moves are at his choice. Maybe it's better to start with a set of strong N-mover positions, say N=6 in case of a 2500 GM, similar to the strong one shown for 8 moves.

These are very interesting "initiative" handicaps, maybe with two small snags. First, it's harder to assess for the general public the value of the handicap, compared to say 2 pawns. Second, if successful for Komodo, one can argue that engines are pretty dumb and materialistic, they have difficulty overcoming material disadvantages, as shown in the previous 2 matches, but they don't care too much about positional asymmetries.

How did you determine that this is the maximum achievement with eight moves? I'm not saying that you are wrong, just wondered how you can be sure. With 9 moves White can get a plus 6, pretty much resignable for Black. If only the final position of the White pieces has to be in the first four ranks rather than the individual moves. White can achieve plus 6 with 8 moves.

No, I should have made it clearer, this is just a guess, then, after trying several more positions, it still was the best. I have no proof whatsoever.

[D]rnbqkbnr/pppppppp/8/8/2BPPB2/2N2N2/PPP1QPPP/2KR3R w kq -

1. e4
2. d4
3. nf3
4. nc3
5. Bc4
6. Bf4
7. Qe2
8. 0-0-0

(Individual moves)

I don't have Komodo to check this postion, but it may well be 6+ already. I don't know if you specified that Bishops can't come to the 4th rank, but if they can it's over. Bxf7+ is threatened...

DB

Yes, indeed, Komodo at +4.67 after 8 minutes on my lowly notebook sees the threat as decisive. 8 moves are too much. Can you find something decisive, say more than +3.60, for 7 moves? I have tried several positions similar to yours after 7 moves with no better than +3.60, but a pawn is lost directly. Would mean roughly 150-170 ELO points per move, so against a 2200 FM 6-7 moves, against 2500 GM 4-5 moves, against 2700 GM 3-4 moves. The problem remains of repeated opening schemes by the human, maybe having forced simplifications and easy draws, if not wins. Still, an interesting handicap, but I am not sure what impression it leaves on the public. Maybe it's better to have a set of "normal" opening N-movers with no decisive tactical shots. But 2 varied pawns (one of them a2 or h2) seem a surer bet against a 2500 GM as the perceived achievement.

EDIT:
The 7-mover
[D]rnbqkbnr/pppppppp/8/8/2BPPB2/2N2N2/PPP1QPPP/R3K2R w KQkq - 0 1

rose to +3.90 after half an hour, with a lost d-pawn. Seems too much again.

[Jesse Gersenson]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

[h1a8]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

[lkaufman]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

Actually these matches and this discussion have caused me to focus on whether Komodo properly balances material vs. positional factors. Based on Kai's tests and the engine evals, it seems that Komodo rather undervalues material in the opening, even though that has been well tuned. For example, if you give White seven or eight free moves at the start (in his own half of the board) but remove his queen's knight, who is really better? Is the engine eval of such starting positions an accurate predictor of the results?

[thekingman]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

The problem is engines are simply not designed to be good at material odds. Any GM could give me a much more competitive game with rook odds than any computer. Humans intuitively understand that in such situations, you need to complicate and attack. Computers do not and will play their best "normal" chess, which is rarely enough to salvage such an already losing position. Move odds allow the engines to do something closer to what they're designed for.

And then there's the nature of the games themselves to consider. With material odds, a successful game by the human involves them doing nothing interesting and just methodically trading pieces. Honestly, who wants to see that? Move odds guarantees that whatever the result, there would be an exciting battle every game.

[h1a8]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

Actually these matches and this discussion have caused me to focus on whether Komodo properly balances material vs. positional factors. Based on Kai's tests and the engine evals, it seems that Komodo rather undervalues material in the opening, even though that has been well tuned. For example, if you give White seven or eight free moves at the start (in his own half of the board) but remove his queen's knight, who is really better? Is the engine eval of such starting positions an accurate predictor of the results?

good insight!

This will increase the understanding and values of the positional sacrifice of a material for the gain of several tempi with the initiative.

[Laskos]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

Actually these matches and this discussion have caused me to focus on whether Komodo properly balances material vs. positional factors. Based on Kai's tests and the engine evals, it seems that Komodo rather undervalues material in the opening, even though that has been well tuned. For example, if you give White seven or eight free moves at the start (in his own half of the board) but remove his queen's knight, who is really better? Is the engine eval of such starting positions an accurate predictor of the results?

On my weak notebook, I was able to confirm that Komodo overvalues during the opening the positional factors compared to material in match outcomes at ultra-fast. With that positional 7-mover evaluated at +3.90, larger than knight b1 handicap, I played time-odds matches, roughly matching the time handicap needed for the knight handicap.

Same time handicap, two matches 100 games each, ultra-fast.

Knight handicapped Komodo:
+30 -46 =24
meaning that the time handicap (ELO handicap) chosen was a bit too low, and the real ELO handicap is a bit larger.

Positional 7-mover eval of -3.90
+89 -6 =5
meaning that the time handicap chosen was way too high, and the real ELO handicap is much smaller.

It translates into that a positional handicap in the opening of 3.90 is much smaller as match outcome ELO goes in ultra-fast than a knight handicap (3.40 or so). I am not sure that this holds at longer time controls, I guess positional factors will count more in that case, and Komodo is not that off in the opening at LTC.

[lkaufman]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

Actually these matches and this discussion have caused me to focus on whether Komodo properly balances material vs. positional factors. Based on Kai's tests and the engine evals, it seems that Komodo rather undervalues material in the opening, even though that has been well tuned. For example, if you give White seven or eight free moves at the start (in his own half of the board) but remove his queen's knight, who is really better? Is the engine eval of such starting positions an accurate predictor of the results?

On my weak notebook, I was able to confirm that Komodo overvalues during the opening the positional factors compared to material in match outcomes at ultra-fast. With that positional 7-mover evaluated at +3.90, larger than knight b1 handicap, I played time-odds matches, roughly matching the time handicap needed for the knight handicap.

Same time handicap, two matches 100 games each, ultra-fast.

Knight handicapped Komodo:
+30 -46 =24
meaning that the time handicap (ELO handicap) chosen was a bit too low, and the real ELO handicap is a bit larger.

Positional 7-mover eval of -3.90
+89 -6 =5
meaning that the time handicap chosen was way too high, and the real ELO handicap is much smaller.

It translates into that a positional handicap in the opening of 3.90 is much smaller as match outcome ELO goes in ultra-fast than a knight handicap (3.40 or so). I am not sure that this holds at longer time controls, I guess positional factors will count more in that case, and Komodo is not that off in the opening at LTC.

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?

[Jesse Gersenson]

...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?

[Uri Blass]

...Still, an interesting handicap, but I am not sure what impression it leaves on the public.

That's the core of these matches - public opinion is that the engine will crush the human. Period. They don't bother the nuances of handicaps. This is confirmed by the voting on the probable outcomes of the FM knight odds match and the first two pawn GM match. Everyone thinks the machine will win. Truth is, the machine gets badly beaten.

Staying with classical odds is preferred as that provides a reference point.

Devising novel odds which have no historic reference is absurd in so far as the value of the odds is undefined. Win, lose or draw the value of the achievement, or failure, is circus-like.

Take regular material odds and figure out what human players are suited to the odds. Start with rook odds.

we should encourage the attempt of innovation. I find move odds very interesting. Exploring further, we could figure out the true value of tempi and why some passive openings are still strong enough to play. I also believe that today's top engines evaluate positions very accurately. So a 3 pawn tempi advantage could be similar to Knights odds.

Anyway I say encourage them and not criticise. After all, there is nothing to lose but boredom. We need our hobby to stay interesting. If we always remained constant then even the classical Knight odds would have been explored thoroughly.

Actually these matches and this discussion have caused me to focus on whether Komodo properly balances material vs. positional factors. Based on Kai's tests and the engine evals, it seems that Komodo rather undervalues material in the opening, even though that has been well tuned. For example, if you give White seven or eight free moves at the start (in his own half of the board) but remove his queen's knight, who is really better? Is the engine eval of such starting positions an accurate predictor of the results?

On my weak notebook, I was able to confirm that Komodo overvalues during the opening the positional factors compared to material in match outcomes at ultra-fast. With that positional 7-mover evaluated at +3.90, larger than knight b1 handicap, I played time-odds matches, roughly matching the time handicap needed for the knight handicap.

Same time handicap, two matches 100 games each, ultra-fast.

Knight handicapped Komodo:
+30 -46 =24
meaning that the time handicap (ELO handicap) chosen was a bit too low, and the real ELO handicap is a bit larger.

Positional 7-mover eval of -3.90
+89 -6 =5
meaning that the time handicap chosen was way too high, and the real ELO handicap is much smaller.

It translates into that a positional handicap in the opening of 3.90 is much smaller as match outcome ELO goes in ultra-fast than a knight handicap (3.40 or so). I am not sure that this holds at longer time controls, I guess positional factors will count more in that case, and Komodo is not that off in the opening at LTC.

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?

Maybe you need to reduce only the score of big positional advantage and not reduce the score of small positional advantage.

You can have a table that tell you that positional advantage below 1 pawn is the same or even slightly bigger but positional advantage above 1 pawn is smaller.

another idea is that maybe huge positional advantage is not so huge when there are many pawns in the board or many pieces in the board.