A "persistent score" flag

Rodolfo Leoni · Post by **Rodolfo Leoni** » Sun Jul 23, 2017 11:54 am

The position at bottom of topic (or similar) often appear from engines games, when I let them play the King's Indian Mar del Plata, out of book. I mean, they start playing after this opening variation:

[pgn][Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[ECO "E99p"]

1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 O-O 6.Be2 e5 7.O-O Nc6 8.d5 Ne7
9.Ne1 Nd7 10.f3 f5 11.g4 Nf6 12.Nd3 *

[/pgn]

[d]r1bq1rk1/ppp1n1bp/3p1np1/3Ppp2/2P1P1P1/2NN1P2/PP2B2P/R1BQ1RK1 b - - 2 12

It's obvious engines cannot think at "strategies", so they'll often don't open lines on the Q side as white, or they'll often don't try to attack kign's castling when playing it as black. Evaluation factors as, for example, space control or pieces mobility can give a wrong score output.

The following position is a draw, but every engine will score it with a little advantage for white. The only "flag" about a possible draw is the "persistent score", +87 cps, since depth 30. And my question is:

Can this search pattern be detected to score such positions as draw?

Thanks.

[d]r1q2rk1/2nbn1b1/1p1p2p1/p1pPp1Pp/P1P1Pp1P/1PNN1P2/2Q2R1K/R3BB2 w - - 0 1

Code: Select all

FEN&#58; r1q2rk1/2nbn1b1/1p1p2p1/p1pPp1Pp/P1P1Pp1P/1PNN1P2/2Q2R1K/R3BB2 w - - 0 1

Stockfish_17071401_x64_modern&#58;
 8/8	00&#58;00	 4k	1,809k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7
 9/9	00&#58;00	 4k	2,134k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Qf1
 10/10	00&#58;00	 5k	1,762k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6
 11/11	00&#58;00	 6k	2,093k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6 6.Qf1
 12/12	00&#58;00	 8k	2,550k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6 6.Qf1 Nb4
 13/15	00&#58;00	 10k	3,276k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6 6.Qf1 Nb4 7.Rd1 Qe7 8.Bb2
 14/16	00&#58;00	 12k	3,903k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6 6.Qf1 Nb4 7.Rd1 Qe7 8.Bb2 Bxb5
 15/16	00&#58;00	 13k	3,354k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Re2 Nc8 4.Qc1 Nc7 5.Bg2 Na6 6.Qf1 Nb4 7.Rd1 Qe7 8.Bb2 Bxb5
 16/18	00&#58;00	 36k	4,537k	+0.92	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Rf7 6.Rd1 Na6 7.Rff1 Qe7 8.Bb2 Nb4
 17/21	00&#58;00	 60k	5,434k	+0.89	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Rf7 6.Rd1 Nxb5 7.axb5 a4 8.Ra1 a3 9.Qf1 Bh8 10.Rfa2
 18/18	00&#58;00	 83k	5,939k	+0.89	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Rf7 6.Rd1 Nxb5 7.axb5 a4 8.Ra1 a3 9.Qf1 Bh8
 19/19	00&#58;00	 128k	6,391k	+0.89	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Rf7 6.Rd1 Nxb5 7.axb5 a4 8.Ra1 a3 9.Qf1 Bh8 10.Rfa2
 20/21	00&#58;00	 200k	6,904k	+0.94	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Rf7 6.Rd1 Na6 7.Qf1 Nb4 8.Bb2 Qe7 9.Nc3 Qd8 10.Rfd2 Ne7 11.Nf2
 21/25	00&#58;00	 492k	6,838k	+0.94	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Rd1 Qe7 7.Bb2 Na7 8.Nc3 Rae8 9.Qf1 Nb4 10.Rfd2 Nc8 11.Nf2 Qd8 12.Bh3
 22/22	00&#58;00	 533k	6,742k	+0.94	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Na7 5.Kg1 Qc8 6.Rc1 Qd8 7.Qd2 Qe7 8.Re1 Nc8 9.Rd1 Qd8 10.Qb2 Nc7 11.Qe2 Nxb5
 23/24	00&#58;00	 620k	6,669k	+0.94	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Na7 5.Kg1 Qc8 6.Rc1 Qd8 7.Qd2 Qe7 8.Re1 Nc8 9.Rd1 Qd8 10.Qb2 Nc7 11.Qe2 Nxb5 12.axb5 a4
 24/25	00&#58;00	 839k	6,312k	+0.92	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Na7 5.Kg1 Qc8 6.Rc1 Qd8 7.Qd2 Nc8 8.Bb2 Nc7 9.Qe1 Nxb5 10.axb5 Ne7 11.Ra1 Kh7 12.Bc3 Qc7 13.Re2
 25/31	00&#58;00	 1,583k	6,408k	+0.93	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Na7 5.Kg1 Qc8 6.Rc1 Qd8 7.Qd2 Nc8 8.Re2 Nc7 9.Qe1 Na6 10.Nf2 Nb4 11.Qf1 Qe7 12.Bh3 Na7 13.Na3
 26/31	00&#58;00	 2,509k	6,351k	+0.89	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Rd1 Nb4 7.Bb2 Qe7 8.Qf1 Na7 9.Nc3 Rae8 10.Rfd2 Rf7 11.Nf2 Nc8 12.Ba3 Na7 13.Bh3 Na6 14.Bxd7 Qxd7
 27/30	00&#58;00	 3,207k	6,388k	+0.93	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Nb4 7.Qf1 Qe7 8.Nc3 Na7 9.Rd2 Rae8 10.Nf2 Qd8 11.Rad1 Qc8 12.Qe2 Qd8 13.Ba3 Rf7 14.Qf1
 28/30	00&#58;00	 3,975k	6,370k	+0.91	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Nb4 7.Rd1 Qe7 8.Nc3 Na7 9.Qf1 Rae8 10.Rfd2 Rf7 11.Ba3 Nc8 12.Nf2 Na7 13.Bh3 Bxh3 14.Qxh3 Qd8
 29/31	00&#58;00	 5,789k	6,245k	+0.93	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Qe7 7.Rd1 Nb4 8.Qf1 Na7 9.Nc3 Rae8 10.Rfd2 Rf7 11.Ba3 Nc8 12.Nf2 Qd8 13.Re2 Na7 14.Bh3 Qc8 15.Bxd7 Qxd7 16.Rdd2
 30/30	00&#58;01	 6,436k	6,218k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Qe7 7.Rd1 Nb4 8.Nc3 Na7 9.Qf1 Rae8 10.Re2 Rf7 11.Red2 Qd8 12.Nf2 Qc8 13.Ba3 Qd8 14.Re1 Qc8 15.Rc1 Qd8
 31/31	00&#58;01	 7,169k	6,191k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Qe7 7.Rd1 Nb4 8.Nc3 Na7 9.Qf1 Rae8 10.Re2 Rf7 11.Red2 Qd8 12.Nf2 Qc8 13.Ba3 Qd8 14.Re1 Qc8 15.Rc1 Qd8 16.Bh3
 32/34	00&#58;01	 10,300k	5,971k	+0.87	1.Nb5 Ne8 2.Bc3 Qd8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Qe7 7.Rd1 Nb4 8.Nc3 Na7 9.Qf1 Rae8 10.Re2 Rf7 11.Red2 Qd8 12.Nf2 Qc8 13.Ba3 Qd8 14.Re1 Qc8 15.Rc1 Qd8 16.Bh3 Qe7 17.Rdd1 Qd8
 33/38	00&#58;16	 84,157k	5,173k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Qf1 Na6 6.Bh3 Rf7 7.Bxd7 Rxd7 8.Bb2 Na7 9.Nc3 Rf7 10.Rd2 Qc8 11.Kg1 Nb4 12.Rad1 Qd7 13.Rh2 Re8 14.Qg2 Rff8 15.Qe2 Nc8 16.Rd2 Re7 17.Rd1 Rff7
 34/36	00&#58;17	 90,852k	5,144k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Qf1 Rf7 6.Bh3 Bxh3 7.Qxh3 Re7 8.Bb2 Na6 9.Kg1 Na7 10.Nc3 Qc8 11.Qf1 Rf7 12.Rd1 Nb4 13.Rh2 Qd7 14.Qe1 Re8 15.Qd2 Rff8 16.Qf2 Nc8 17.Qf1 Rf7 18.Qg2 Rff8
 35/36	00&#58;23	 118,417k	5,037k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Qf1 Rf7 6.Bh3 Bxh3 7.Qxh3 Re7 8.Bb2 Na6 9.Kg1 Na7 10.Nc3 Qc8 11.Qf1 Rf7 12.Rd1 Nb4 13.Rh2 Qd7 14.Qe1 Re8 15.Qd2 Rff8 16.Qf2 Nc8 17.Qf1 Rf7 18.Rdd2 Qe7
 36/39	00&#58;36	 175,888k	4,839k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Nb4 7.Rff1 Na7 8.Nc3 Qb7 9.Nf2 Rae8 10.Rad1 Nc8 11.Nb5 Qb8 12.Qd2 Na7 13.Nc3 Nc8 14.Bh3 Bxh3 15.Nxh3 Qb7 16.Nf2 Ne7 17.Ra1 Qd7 18.Nb5 Na6 19.Rac1 Rd8 20.Rcd1 Ra8 21.Rc1
 37/47	02&#58;30	 587,711k	3,898k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rac1 Nc8 11.Qd2 Kh7 12.Rcd1 Qc7 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Qb7 16.Nf2 Qd7 17.Kh2 Ne7 18.Kg2 Rd8 19.Rde1 Na6 20.Rh1 Ra8
 38/41	02&#58;37	 607,641k	3,866k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Ne7 16.Nf2 Qd8 17.Kg2 Qd7 18.Kh2 Na6 19.Kg1 Nc8 20.Kg2 Ne7 21.Qe2
 39/41	02&#58;54	 662,266k	3,792k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Ne7 16.Nf2 Qd8 17.Kg2 Qd7 18.Kh2 Na6 19.Kg1 Nc8 20.Kg2 Ne7 21.Qe2
 40/42	03&#58;03	 690,308k	3,763k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Ne7 16.Nf2 Qd8 17.Kg2 Qd7 18.Kh2 Na6 19.Kg1 Nc8 20.Rc1 Ne7 21.Bb2 Nb4
 41/43	03&#58;39	 804,006k	3,667k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Ne7 16.Nf2 Qd8 17.Kg2 Qd7 18.Kh2 Na6 19.Kg1 Nc8 20.Bb2 Nb4 21.Qe2 Na7 22.Nc3 Nc8 23.Ba3
 42/44	04&#58;06	 892,703k	3,615k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Bh3 Bxh3 15.Nxh3 Ne7 16.Nf2 Qd8 17.Kg2 Qd7 18.Kh2 Na6 19.Kg1 Nc8 20.Bb2 Nb4 21.Qe2 Na7 22.Nc3 Nc8
 43/44	04&#58;23	 945,687k	3,584k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Qe2 Kh7 15.Bb2 Kg8 16.Nc3 Ne7 17.Qd2 Nc8 18.Rfe1 Kh7 19.Ba3 Kg8 20.Bh3 Bxh3 21.Nxh3 Qb7 22.Nf2 Qd7
 44/45	06&#58;28	 1,342,530k	3,457k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Qe2 Kh7 15.Bb2 Kg8 16.Nc3 Ne7 17.Qd2 Nc8 18.Nd3 Qc7 19.Nb5 Qb7 20.Nf2 Qb8 21.Nc3 Qb7 22.Ba3 Qb8 23.Bh3 Bxh3
 45/45	09&#58;50	 1,979,736k	3,351k	+0.87	1.Nb5 Ne8 2.Bc3 Qb8 3.Qe2 Nc8 4.Bg2 Nc7 5.Kg1 Na6 6.Bb2 Na7 7.Nc3 Qb7 8.Rff1 Rae8 9.Nf2 Nb4 10.Rad1 Ra8 11.Qd2 Rae8 12.Ba3 Nc8 13.Nb5 Qb8 14.Qe2 Kh7 15.Bb2 Kg8 16.Nc3 Ne7 17.Qd2 Nc8 18.Nd3 Qc7 19.Nb5 Qb7 20.Nf2 Qb8 21.Nc3 Qb7 22.Ba3 Qb8 23.Bh3 Bxh3

zullil · Post by **zullil** » Sun Jul 23, 2017 12:16 pm

Rodolfo Leoni wrote:
The only "flag" about a possible draw is the "persistent score", +87 cps, since depth 30. And my question is:

Can this search pattern be detected to score such positions as draw?

Thanks.

The pattern only appears after the search is made. Suppose the position in question appeared at a leaf in a search from an earlier position. How would the engine know that, with further searching, the evaluations of that leaf position would eventually show no progress? Hence the need for a static evaluation process.

Rodolfo Leoni · Post by **Rodolfo Leoni** » Sun Jul 23, 2017 12:25 pm

zullil wrote:
Rodolfo Leoni wrote:
The only "flag" about a possible draw is the "persistent score", +87 cps, since depth 30. And my question is:

Can this search pattern be detected to score such positions as draw?

Thanks.
The pattern only appears after the search is made. Suppose the position in question appeared at a leaf in a search from an earlier position. How would the engine know that, with further searching, the evaluations of that leaf position would eventually show no progress? Hence the need for a static evaluation process.

Being not a chess programmer (except for some easy VB tools), I've a lot of doubts too. I'm trying to better understand the process, which looks very similar to the one of the "Easy position, for humans". Engine tries to avoid 3 fold repetitions while on search, and I wonder if it can be detected....

But I accept the fact my understanding of the problem is very limited.

hgm · Post by **hgm** » Sun Jul 23, 2017 12:37 pm

A much more sensible approach would be to have the engine recognize the position as drawish in the static evaluation. It doesn't seem to be very difficult in this case to recognize some drawish aspects. Such as the Pawn structure, penetrable only in one place, which can be abundantly defended. This would be pretty cheap, as Pawn evaluations are hashed, so that even quite complex analysis would not have measurable impact on speed. Detecting such a drawishness indicator could make the engine divide the evaluation score by 2, which would probably be enough to reject it compared to more promising lines of play.

Rodolfo Leoni · Post by **Rodolfo Leoni** » Sun Jul 23, 2017 12:46 pm

hgm wrote:A much more sensible approach would be to have the engine recognize the position as drawish in the static evaluation. It doesn't seem to be very difficult in this case to recognize some drawish aspects. Such as the Pawn structure, penetrable only in one place, which can be abundantly defended. This would be pretty cheap, as Pawn evaluations are hashed, so that even quite complex analysis would not have measurable impact on speed. Detecting such a drawishness indicator could make the engine divide the evaluation score by 2, which would probably be enough to reject it compared to more promising lines of play.

Looks brilliant. If engines can recognize it, then they'd avoid to close the position. At least, playing white. But black too could find opponent has the best chances on the Q side, and it possibly could choose not to close the K side...

Thanks.

A "persistent score" flag

A "persistent score" flag

Re: A "persistent score" flag

Re: A "persistent score" flag

Re: A "persistent score" flag

Re: A "persistent score" flag