Komodo knight odd match vs FM Victor Bolzoni : report

[lkaufman]

My bad, my english is not always good.
Will Komodo be better without book at the opening ?

I don't think it matters much whether or not Komodo has a book for the first game at each handicap. Since there is no theory on these handicaps, I probably couldn't do much better than just have Komodo pre-select its own opening moves, which would save a little clock time so might be worth five elo or so. But once a game has been played at a given handicap, a book is helpful for avoiding any preparation the opponent might do in case Komodo simply repeated its play.

[Laskos]

Thanks, this is very helpful and pretty impressive, nearly 140 elo swing from the shortest to longest tc tested. We're trying to verify the principle in general; early results are confirming but samples still small.

Did you get some conclusive results? I tested now on my desktop several similar fairly "equal" openings, knight odds, positionally skewed, generally they tend to behave like what we saw with that position, some few others do not show much progress with time control. Most of them seem under-valued for attacker by Komodo after 10 minute search on 4 cores. Here are examples of 7-movers:

1/

[d]rnbqkbnr/pppppppp/8/8/2BPPB2/5N2/PPP1QPPP/2KR3R w kq -

Code: Select all

 28.01	 4:41 	-1.31++	1.d5 (1.574.732.561) 5598
 28.01	 4:55 	-1.24++	1.d5 (1.654.944.286) 5608
 28.01	 5:22 	-1.14++	1.d5 (1.809.389.134) 5610
 28.01	 5:35 	-1.22--	1.d5 c6 (1.883.816.946) 5619
 28.01	 7:37 	-1.10++	1.d5 (2.572.388.538) 5617
 28.01	 7:44 	-1.11 	1.d5 c6 2.Rhe1 b5 3.Bb3 Qb6 4.Ne5 d6 5.Nxc6 Nxc6 6.dxc6 Be6 7.e5 d5 8.Rd3 Qxc6 9.Red1 Rd8 10.Qf3 a6 11.Rxd5 Rxd5 12.Bxd5 Bxd5 13.Rxd5 Qc8 14.Rd3 (2.606.600.482) 5616
 29.01	 8:34 	-1.18--	1.d5 c6 (2.895.134.770) 5622
 29.01	 9:01 	-1.11++	1.d5 (3.047.421.588) 5631
 29.01	10:00 	-1.18--	1.d5 c6 (3.380.996.604) 5631
 29.01	11:20 	-1.09++	1.d5 (3.822.825.704) 5620
 29.01	12:25 	-1.19 	1.d5 c6 2.Ne5 c5 3.Nf3 e6 4.h4 Qb6 5.h5 Be7 6.Kb1 Nf6 7.h6 g6 8.Rd3 a6 9.Ng5 O-O 10.d6 Bd8 11.e5 Nd5 12.Bxd5 exd5 13.Rxd5 Qa7 14.Ne4 (4.187.934.693) 5620

Code: Select all

1.5''+0.015''
Games        :    200 (finished)

White Wins   :     50 (25.0 %)
Black Wins   :    125 (62.5 %)
Draws        :     25 (12.5 %)
Unfinished   :      0

White Perf.  : 31.2 %
Black Perf.  : 68.8 %

6''+0.06''
Games        :    200 (finished)

White Wins   :     72 (36.0 %)
Black Wins   :     79 (39.5 %)
Draws        :     49 (24.5 %)
Unfinished   :      0

White Perf.  : 48.2 %
Black Perf.  : 51.8 %

24''+0.24''
Games        :    100 (finished)

White Wins   :     37 (37.0 %)
Black Wins   :     27 (27.0 %)
Draws        :     36 (36.0 %)
Unfinished   :      0

White Perf.  : 55.0 %
Black Perf.  : 45.0 %

2/

[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N5/PPP1QPPP/2KR3R w kq -

Code: Select all

 28.01	 4:26 	-0.52 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a5 5.Be3 Qd8 6.Ne4 a4 7.a3 Qa5 8.f4 b5 9.Bd2 Qa6 10.Bd3 Qb7 11.Nc5 Qb6 12.Be3 Qd8 13.f5 exf5 14.Bxf5 (1.517.769.448) 5687
 29.01	 4:35 	-0.59--	1.Kb1 c6 (1.566.678.196) 5684
 29.01	 5:23 	-0.52++	1.Kb1 (1.843.720.278) 5702
 29.01	 5:25 	-0.59--	1.Kb1 c6 (1.859.527.697) 5704
 29.01	 5:39 	-0.64 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a5 5.Be3 Qd8 6.Ne4 a4 7.a3 b5 8.Bd3 Qa5 9.Ng5 Nh6 10.Nxh7 b4 11.Bxh6 bxa3 12.Bg5 axb2 13.Kxb2 Ba6 14.Nxf8 (1.941.234.566) 5720
 30.01	 6:06 	-0.57++	1.Kb1 (2.097.852.927) 5723
 30.01	 8:40 	-0.50++	1.Kb1 (2.998.659.098) 5756
 30.01	 8:49 	-0.57--	1.Kb1 c6 (3.048.940.692) 5756
 30.01	 9:09 	-0.72--	1.Kb1 c6 (3.161.862.788) 5754
 30.01	10:02 	-0.60++	1.Kb1 (3.471.227.398) 5764
 30.01	13:50 	-0.46 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a6 5.Be3 c5 6.Ne4 Qc6 7.Bxc5 b6 8.Bd4 b5 9.Bd3 Bb7 10.f4 Qc8 11.Be3 Nh6 12.h3 Bd5 13.Bc5 Nf5 14.Bf2 (4.777.647.137) 5753

Code: Select all

1.5''+0.15''
Games        :    200 (finished)

White Wins   :     88 (44.0 %)
Black Wins   :     82 (41.0 %)
Draws        :     30 (15.0 %)
Unfinished   :      0

White Perf.  : 51.5 %
Black Perf.  : 48.5 %

6''+0.06''
Games        :    200 (finished)

White Wins   :     67 (33.5 %)
Black Wins   :     83 (41.5 %)
Draws        :     50 (25.0 %)
Unfinished   :      0

White Perf.  : 46.0 %
Black Perf.  : 54.0 %

24''+0.24''
Games        :    100 (finished)

White Wins   :     41 (41.0 %)
Black Wins   :     40 (40.0 %)
Draws        :     19 (19.0 %)
Unfinished   :      0

White Perf.  : 50.5 %
Black Perf.  : 49.5 %

[lkaufman]

Thanks, this is very helpful and pretty impressive, nearly 140 elo swing from the shortest to longest tc tested. We're trying to verify the principle in general; early results are confirming but samples still small.

Did you get some conclusive results? I tested now on my desktop several similar fairly "equal" openings, knight odds, positionally skewed, generally they tend to behave like what we saw with that position, some few others do not show much progress with time control. Most of them seem under-valued for attacker by Komodo after 10 minute search on 4 cores. Here are examples of 7-movers:

1/

[d]rnbqkbnr/pppppppp/8/8/2BPPB2/5N2/PPP1QPPP/2KR3R w kq -

Code: Select all

 28.01	 4:41 	-1.31++	1.d5 (1.574.732.561) 5598
 28.01	 4:55 	-1.24++	1.d5 (1.654.944.286) 5608
 28.01	 5:22 	-1.14++	1.d5 (1.809.389.134) 5610
 28.01	 5:35 	-1.22--	1.d5 c6 (1.883.816.946) 5619
 28.01	 7:37 	-1.10++	1.d5 (2.572.388.538) 5617
 28.01	 7:44 	-1.11 	1.d5 c6 2.Rhe1 b5 3.Bb3 Qb6 4.Ne5 d6 5.Nxc6 Nxc6 6.dxc6 Be6 7.e5 d5 8.Rd3 Qxc6 9.Red1 Rd8 10.Qf3 a6 11.Rxd5 Rxd5 12.Bxd5 Bxd5 13.Rxd5 Qc8 14.Rd3 (2.606.600.482) 5616
 29.01	 8:34 	-1.18--	1.d5 c6 (2.895.134.770) 5622
 29.01	 9:01 	-1.11++	1.d5 (3.047.421.588) 5631
 29.01	10:00 	-1.18--	1.d5 c6 (3.380.996.604) 5631
 29.01	11:20 	-1.09++	1.d5 (3.822.825.704) 5620
 29.01	12:25 	-1.19 	1.d5 c6 2.Ne5 c5 3.Nf3 e6 4.h4 Qb6 5.h5 Be7 6.Kb1 Nf6 7.h6 g6 8.Rd3 a6 9.Ng5 O-O 10.d6 Bd8 11.e5 Nd5 12.Bxd5 exd5 13.Rxd5 Qa7 14.Ne4 (4.187.934.693) 5620

Code: Select all

1.5''+0.015''
Games        :    200 (finished)

White Wins   :     50 (25.0 %)
Black Wins   :    125 (62.5 %)
Draws        :     25 (12.5 %)
Unfinished   :      0

White Perf.  : 31.2 %
Black Perf.  : 68.8 %

6''+0.06''
Games        :    200 (finished)

White Wins   :     72 (36.0 %)
Black Wins   :     79 (39.5 %)
Draws        :     49 (24.5 %)
Unfinished   :      0

White Perf.  : 48.2 %
Black Perf.  : 51.8 %

24''+0.24''
Games        :    100 (finished)

White Wins   :     37 (37.0 %)
Black Wins   :     27 (27.0 %)
Draws        :     36 (36.0 %)
Unfinished   :      0

White Perf.  : 55.0 %
Black Perf.  : 45.0 %

2/

[d]rnbqkbnr/pppppppp/8/8/2BPPB2/2N5/PPP1QPPP/2KR3R w kq -

Code: Select all

 28.01	 4:26 	-0.52 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a5 5.Be3 Qd8 6.Ne4 a4 7.a3 Qa5 8.f4 b5 9.Bd2 Qa6 10.Bd3 Qb7 11.Nc5 Qb6 12.Be3 Qd8 13.f5 exf5 14.Bxf5 (1.517.769.448) 5687
 29.01	 4:35 	-0.59--	1.Kb1 c6 (1.566.678.196) 5684
 29.01	 5:23 	-0.52++	1.Kb1 (1.843.720.278) 5702
 29.01	 5:25 	-0.59--	1.Kb1 c6 (1.859.527.697) 5704
 29.01	 5:39 	-0.64 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a5 5.Be3 Qd8 6.Ne4 a4 7.a3 b5 8.Bd3 Qa5 9.Ng5 Nh6 10.Nxh7 b4 11.Bxh6 bxa3 12.Bg5 axb2 13.Kxb2 Ba6 14.Nxf8 (1.941.234.566) 5720
 30.01	 6:06 	-0.57++	1.Kb1 (2.097.852.927) 5723
 30.01	 8:40 	-0.50++	1.Kb1 (2.998.659.098) 5756
 30.01	 8:49 	-0.57--	1.Kb1 c6 (3.048.940.692) 5756
 30.01	 9:09 	-0.72--	1.Kb1 c6 (3.161.862.788) 5754
 30.01	10:02 	-0.60++	1.Kb1 (3.471.227.398) 5764
 30.01	13:50 	-0.46 	1.Kb1 c6 2.d5 Qb6 3.e5 e6 4.d6 a6 5.Be3 c5 6.Ne4 Qc6 7.Bxc5 b6 8.Bd4 b5 9.Bd3 Bb7 10.f4 Qc8 11.Be3 Nh6 12.h3 Bd5 13.Bc5 Nf5 14.Bf2 (4.777.647.137) 5753

Code: Select all

1.5''+0.15''
Games        :    200 (finished)

White Wins   :     88 (44.0 %)
Black Wins   :     82 (41.0 %)
Draws        :     30 (15.0 %)
Unfinished   :      0

White Perf.  : 51.5 %
Black Perf.  : 48.5 %

6''+0.06''
Games        :    200 (finished)

White Wins   :     67 (33.5 %)
Black Wins   :     83 (41.5 %)
Draws        :     50 (25.0 %)
Unfinished   :      0

White Perf.  : 46.0 %
Black Perf.  : 54.0 %

24''+0.24''
Games        :    100 (finished)

White Wins   :     41 (41.0 %)
Black Wins   :     40 (40.0 %)
Draws        :     19 (19.0 %)
Unfinished   :      0

White Perf.  : 50.5 %
Black Perf.  : 49.5 %

Although your first position strongly supports the hypothesis that more time favors the attacker, the second one does not. I guess you would need a lot of positions to be sure. What we have done is to determine whether increasing the weight on positional terms might help with longer time limits than what we usually use. What we've found is that a 3% increase does appear to help at 2' + 1" while being slightly negative at 20" +.1". So probably next release will have this 3% increase since we obviously care a lot more about the longer tc. 3% isn't very much (6% did worse at both levels) but anyway if we gain a couple elo from it we can thank you for pointing us in this direction. We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

[cdani]

We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Do you mean for example what I do in Andscacs, that it increases slighty the material imbalance bonus if there is big difference in mobility?

Code: Select all

//y = difference of mobility (from mg)
//z = compensation bonus, for mg and eg
			if (abs(y) > 50) 
				z = 15;
			if (abs(y) > 100)
				z = 25;
			if (abs(y) > 150)
				z = 45;

[cdani]

We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Do you mean for example what I do in Andscacs, that it increases slighty the material imbalance bonus if there is big difference in mobility?

Code: Select all

//y = difference of mobility (from mg)
//z = compensation bonus, for mg and eg
			if (abs(y) > 50) 
				z = 15;
			if (abs(y) > 100)
				z = 25;
			if (abs(y) > 150)
				z = 45;

Now fires when there is material difference less than a knight. Needs to be improved and tested for bigger material differences, but already was a win.

[Laskos]

Although your first position strongly supports the hypothesis that more time favors the attacker, the second one does not. I guess you would need a lot of positions to be sure. What we have done is to determine whether increasing the weight on positional terms might help with longer time limits than what we usually use. What we've found is that a 3% increase does appear to help at 2' + 1" while being slightly negative at 20" +.1". So probably next release will have this 3% increase since we obviously care a lot more about the longer tc. 3% isn't very much (6% did worse at both levels) but anyway if we gain a couple elo from it we can thank you for pointing us in this direction. We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Yes, I tested 8-9 unbalanced openings, 6-7 were like the first one, with improvement at longer TC, 2 without. I presented the examples of these 2 different cases. Not sure what's the matter, to me they all looked similar in disbalance. I tried to avoid positions with clear tactical gains for the attacker. No need to mention me for this almost non-improvement.

[lkaufman]

We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Do you mean for example what I do in Andscacs, that it increases slighty the material imbalance bonus if there is big difference in mobility?

Code: Select all

//y = difference of mobility (from mg)
//z = compensation bonus, for mg and eg
			if (abs(y) > 50) 
				z = 15;
			if (abs(y) > 100)
				z = 25;
			if (abs(y) > 150)
				z = 45;

That seems backward; you are saying that material is more important when opposed by mobility, but Kai's results suggest the opposite, that material is more important when it is NOT opposed by positional considerations. Also, I don't understand how z interacts with material imbalance; is it a percentage increase or something like that?

[cdani]

We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Do you mean for example what I do in Andscacs, that it increases slighty the material imbalance bonus if there is big difference in mobility?

Code: Select all

//y = difference of mobility (from mg)
//z = compensation bonus, for mg and eg
			if (abs(y) > 50) 
				z = 15;
			if (abs(y) > 100)
				z = 25;
			if (abs(y) > 150)
				z = 45;

That seems backward; you are saying that material is more important when opposed by mobility, but Kai's results suggest the opposite, that material is more important when it is NOT opposed by positional considerations. Also, I don't understand how z interacts with material imbalance; is it a percentage increase or something like that?

The function changes the sign depending on the color. You cannot view the effect with the part of the code I put here. I put it like this just to show the values I use to give an idea of the magnitude. So in fact I say the same as Kai. And z is the value of compensation in centipawns.

So if the side with less material has more than 150 cp more of mobility, it will have an added bonus of (eg/mg) (45/45) cp.

[lkaufman]

We've tried a couple ideas out to capture the principle that positional factors seem more important as compensation for material than if they are just "free", but so far nothing like that has helped. Any ideas would be welcome.

Do you mean for example what I do in Andscacs, that it increases slighty the material imbalance bonus if there is big difference in mobility?

Code: Select all

//y = difference of mobility (from mg)
//z = compensation bonus, for mg and eg
			if (abs(y) > 50) 
				z = 15;
			if (abs(y) > 100)
				z = 25;
			if (abs(y) > 150)
				z = 45;

That seems backward; you are saying that material is more important when opposed by mobility, but Kai's results suggest the opposite, that material is more important when it is NOT opposed by positional considerations. Also, I don't understand how z interacts with material imbalance; is it a percentage increase or something like that?

The function changes the sign depending on the color. You cannot view the effect with the part of the code I put here. I put it like this just to show the values I use to give an idea of the magnitude. So in fact I say the same as Kai. And z is the value of compensation in centipawns.

So if the side with less material has more than 150 cp more of mobility, it will have an added bonus of (eg/mg) (45/45) cp.

Okay, that makes more sense. What was your rationale for doing this? Did you notice the same phenomenon as Kai, or did you arrive at this some other way? Also, suppose one side is just down a tiny bit of material, say bishop for knight? Then this would work poorly. Or are you just defining material by integer multiples of a pawn (1-3-3-5-9 or perhaps 1-4-4--6-12)?

[cdani]

Okay, that makes more sense. What was your rationale for doing this? Did you notice the same phenomenon as Kai, or did you arrive at this some other way?

Was seeing chess games of Andscacs and by own chess concept. If you see a typical position where there is a side with bad development you know that is probably than the other side can hold the position even with some material less. And then I just tried in Andscacs some related ideas.

Also, suppose one side is just down a tiny bit of material, say bishop for knight? Then this would work poorly. Or are you just defining material by integer multiples of a pawn (1-3-3-5-9 or perhaps 1-4-4--6-12)?

The material is also in cp. The answer is simple, I did not improved this idea Is one more in the list that is pending to be worked seriously.