Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Opening selection for the TCEC 10 superfinal
- - By Venator (Silver) [nl] Date 2017-11-18 11:06 Edited 2017-11-18 16:21 Upvotes 2
Here is some information about the opening lines I selected for the TCEC 10 superfinal:

* Average number of moves per position: 7.44
* Average draw rate: around 7%-8% lower compared to big databases, with the purpose of (slightly) increasing the probability of more wins
* First position is the start position
* The other 49 positions have a length between 3 moves (shortest) and 15 moves (longest)
* The book exits are typically between +0.10 and +0.50, with a few exceptions
* There are a few positions with a slight black edge, to increase the possibility of a black win
* All positions were run through a quick check using the best available engines, to make sure that: A) there were no multiple 0.00 scores, and B) there were no mass exchanges making the game possibly less interesting

What did I do differently compared to the TCEC 9 superfinal:

* I have tried to avoid too extreme book exits and too one sided lines
* I paid more attention to the draw rate, instead of the book exit score
* I accepted lower book exit scores, if in my opinion the position contains enough play to get an interesting game

What kind of lines will we see in the TCEC 10 superfinal:

* All important openings will be played
* There are sharp lines vs positional lines
* There are gambits
* There are SALC-like positions (SALC = "short and long castling")
* There are HERT-like positions (HERT = a testset made by Thomas Zipproth)
* There are a couple of "fun" lines

I am looking forward to an exciting TCEC 10 superfinal, with lots of interesting games and a reasonable amount of wins. Have fun!
Parent - - By brunjes (**) [us] Date 2017-11-18 12:57
Sounds very interesting! Thanks for the info and thanks for your work on TCEC. It's by far the most interesting computer chess event of each year!
Parent - By Venator (Silver) [nl] Date 2017-11-18 15:10
Thanks a lot for your positive words!
Parent - By Nelson Hernandez (Gold) [us] Date 2017-11-22 13:14
Let me second that!
Parent - - By gsgs (**) [de] Date 2017-11-19 01:14 Upvotes 1
not so easy to reduce the drawrate without imbalances,
Why are those excluded ?
You may need an eval of 0.70 at this level to get
a 50% winning chance.
Parent - - By user923005 (****) [us] Date 2017-11-19 03:09
There are openings that have a low ce, have a winning expectation based on stats of close to zero bias, and which have large numbers of wins and losses (obviously wins and losses for either side are about the same number) and a small number of draws.

I attempted an opening set like that called "Fair and Decisive"
Parent - - By gsgs (**) [de] Date 2017-11-19 04:27
I tried to find those.
Best so far (120+1,tablet)
was Komodos's queen sac exf3 from stage 1

1.d4 f5 2.g3 Nf6 3.Bg2 d5 4.Nf3 e6 5.c4 Bd6 6.O-O c6 7.Nc3 O-O 8.Qc2 Ne4 9.Nxe4 dxe4 10.Bg5 exf3

white performance = 49.9%
drawrate = 64.5%
47% decisive (white,black) pairs

1000 games asmfish vs. asmfish 120+1 , x86 tablet with 1800kn/s in the startposition
Parent - By gsgs (**) [de] Date 2017-11-20 05:20 Edited 2017-11-20 07:45
this opening  is actually running at TCEC in a game Stockfish-Komodo

http://tcec.chessdom.com/live.php

evals : +0.87(SF) and -0.00(KO) after move 11  --- +0.32,+0.64
evals : +0.34(SF) and -0.75(KO) after move 19  ---
evals : -0.32(SF) and -1.19(KO) after move 27
evals : -0.48(SF) and -1.41(KO) after move 36
evals : -1.07(SF) and -1.97(KO) after move 45
evals : -1.97(SF) and -2.29(KO) after move 54
evals : -4.63(SF) and -5.74(KO) after move 63

Qd3 - first game , no games in my 2000set  (too bad, the "bonus" game wasn't stored
+111,=316,-076  with that second game
Parent - - By Venator (Silver) [nl] Date 2017-11-19 04:40
Imbalances are not excluded. Besides, you want to avoid two easy 1-0 1-0 (or 0-1 0-1) results, which are quite likely if you choose too extreme book exits. We had some examples of this in the TCEC 9 superfinal.
Parent - - By gsgs (**) [de] Date 2017-11-19 06:35
yes, any opening with a whitescore of ~70% (or 30%) should work well.
When you want ~50% whitescore, that's where it gets difficult.
Parent - By Venator (Silver) [nl] Date 2017-12-08 05:00
I have made an excel sheet with the opening stats of the TCEC 10 superfinal. Seel the attachment below.
Parent - - By MarshallArts (***) [us] Date 2017-11-20 00:03

> not so easy to reduce the drawrate without imbalances,
> Why are those excluded ?
> You may need an eval of 0.70 at this level to get
> a 50% winning chance.


Agreed.

Having such a low threshold (0.50-0.70) will likely exclude many/most King's Indians,
which the engines wrongly evaluate. Why be so rigid? E98-99 Mar del Plata lines
should not be subject to such a ridiculous threshold.

Other than this, I applaud Venator's efforts.

:neutral:
Parent - - By Venator (Silver) [nl] Date 2017-11-20 15:47
Why be so rigid? E98-99 Mar del Plata lines should not be subject to such a ridiculous threshold.

If you read my posting above, I said the threshold was used "with a few exceptions".

Very onesided lines is something we definitely want to avoid. Because it increases the chance of very lopsided 1-0 1-0 results. Two games in which one side wins easily, are IMO uncalled for. They are just as uninteresting as two very short and boring draws.

The TCEC 9 superfinal and stage 2 of TCEC 10 proved that you can have far less extreme lines to get a reasonable amount of decisive results. I.e. the 3 wins amongst the top 3 in stage 2 that ended yesterday, were in the Queen's Indian (twice) and the Ruy Lopez.

Regarding the KI Mar del Plata: T. Poppins played a couple of matches on CCC using my Mar del Plata testsuites and the draw rate was still very high. Even 93% with Mar del Plata II. A match with my Twenty Gambits suite produced 85% draws and a match using my King's Gambit testsuite led to a 72% draw rate.
Parent - By MarshallArts (***) [us] Date 2017-11-21 05:07
My bad for missing the 'few exceptions' part.

I still think that the too-many-draws problem is primarily a function of the closeness of the engines ratings-wise.
Choosing the wrong openings for testing is a separate problem.

As you rightly said, the one in charge of picking the openings has to know how to avoid both drawish and lopsided lines,
and go beyond the simple engine evaluation score for his selection criteria.

:fat:
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-26 21:54

>Agreed.


>Having such a low threshold (0.50-0.70) will likely exclude many/most King's Indians,
>which the engines wrongly evaluate. Why be so rigid? E98-99 Mar del Plata lines
>should not be subject to such a ridiculous threshold.


>Other than this, I applaud Venator's efforts.


MarshallArts,
please check your PM box.
- - By Dr.X (Gold) Date 2017-11-19 16:54
Well, bud, we don't see eye to eye on a great deal of issues! But when it comes to the development of a 'Opening Tournament Book' no one can hold a candle to you! It is easy to see you put a great deal of effort into making this come out all-inclusive. That is an Upvote!
Parent - By Venator (Silver) [nl] Date 2017-11-19 18:30
Thanks!
- By gsgs (**) [de] Date 2017-11-21 06:00 Edited 2017-11-21 09:36
can someone collect drudru and Erdo evals of the openings, please ? (when available)
- - By Hamster (**) [ch] Date 2017-11-23 11:37
In which order are the openings played? Would be nice to have the two engines play one opening after another with alternating colours but it seems quite random to me. In any case, I am surprised about Houdini's performance.
Bogo-Indian: 4.Bd2 Be7 5.g3 d5 6.Bg2 O-O 7.O-O c6 8.Qc2 b6
French: MacCutcheon, Lasker, 7.bxc3
Sicilian: Taimanov, 6.Be3 a6 7.Qd2
Sicilian: Taimanov, 6.Be3 a6 7.Qd2
King's Indian: Fianchetto, Classical, 9.h3 Qb6 10.c5
King's Indian: Fianchetto, Classical, 9.h3 Qb6 10.c5
Spanish: Closed, Smyslov, 12.Nf1 Bd7
Spanish: Closed, Smyslov, 12.Nf1 Bb7 13.Ng3 Na5 14.Bc2 Nc4
Semi-Slav: Moscow, 6.Bh4 Gambit, 7.e4
Semi-Slav: Moscow, 6.Bh4 Gambit, 7.e4
French: Advance, Euwe, 6.a3
French: Advance, Euwe, 6.a3
Budapest: Rubinstein, 4...Bb4+
Budapest: 3...Ng4 4.e3
Caro-Kann: Tartakower, 6.c3
Parent - By free Willy (*) Date 2017-11-23 15:10 Upvotes 1
In the first two games the engines played from the starting position and the next openings were played alternately
cheers
- By gsgs (**) [de] Date 2017-11-28 09:44 Edited 2017-11-28 09:47
evals of openings so far , compared with last season :

50 and 22 openings sorted by absolute value of evals (average of the 2 evals from the 2 engines)

http://magictour.free.fr/TCSFOP.GIF
- By gsgs (**) [de] Date 2017-12-13 12:20 Edited 2017-12-13 13:01
1000 games for each opening at 0.5s per move , x86 tablet , CfA ,
asmFish -  asmFish

sorted by column L



   A   B    C      D      E  F   G   H   I    J  K   L   M   N    O      P
-----------------------------------------------------------------------------------------
522.0,406,605.0,tcect24, 34,112,180,124,50 , 13,082,169,154,82 , 11 , Pirc: 4.Be3, 150 Attack   , T07,d09
427.5,421,543.5,tcect49, 67,155,155,102,21 , 23,112,170,145,50 , 01 , Mar del Plata             , T08,d10
552.0,502,542.0,tcect34, 20,096,180,168,36 , 19,113,175,151,42 , 13 , Najdorf, 6.Be3 e5 7.Nb3   , T24     253
587.5,495,488.5,tcect06, 10,090,168,179,53 , 30,141,176,128,25 , 08 , French - Advance          , T09     275
552.0,510,511.0,tcect35, 14,102,191,152,41 , 31,117,183,137,32 , 19 , English:5.e4 Nb4 6.Bc4    , T31,d09
604.0,472,549.0,tcect29, 11,077,169,179,64 , 20,100,183,156,41 , 03 , Benoni: Four Pawns        ,.I01,d09
631.5,475,556.5,tcect50, 08,054,172,199,67 , 13,105,186,148,48 , 04 , Dragon, Yugoslav          ,,T20,d10
516.5,507,522.5,tcect10, 25,121,186,132,36 , 27,109,188,144,32 , 02 , Sicilian - Richter-Rauzer , T41     369
332.0,440,481.0,tcect28, 91,213,144,045,07 , 31,139,189,119,22 , 19 , Vienna: Steinitz Gambit   ,-T09,d10
478.5,487,506.5,tcect02, 42,131,183,116,20 , 32,116,190,131,31 , 24 , Sicilian - Taimanov       , T09     278

650.5,559,619.5,tcect43, 03,040,166,235,56 , 10,122,201,153,14 , 04 , QGD: Alatortsev, 5.Bf4    , T17,d10
529.5,509,548.5,tcect47, 26,104,196,133,41 , 23,084,203,153,37 , 21 , Semi-Slav                 , T41,d09
458.5,549,499.5,tcect09, 34,145,207,098,16 , 31,114,203,129,23 , 07 , QGD - exchange            , T09     269
656.5,544,533.0,tcect32, 03,035,172,227,63 , 11,107,207,155,20 , 05 , Spanish: Bird's, 5.O-O Bc5,-todo
528.0,578,476.0,tcect16, 15,106,215,136,28 , 23,146,211,096,24 , 25 , French: Chatard           , T36     261
695.5,476,525.0,tcect33, 01,027,151,223,98 , 11,116,212,134,27 , 19 , Neo-Benko, 4.Bg5           -
681.0,508,496.0,tcect30, 00,030,163,222,85 , 20,125,213,127,15 , 17 , French: 3.Nc3 Nc6          -
611.5,575,493.5,tcect03, 07,052,199,195,47 , 24,117,218,130,11 , 09 , KingsIndian - g3
624.0,582,547.0,tcect22, 04,050,191,204,51 , 07,093,219,161,20 , 16 , Caro-Kann: Advance,c5
578.0,592,524.0,tcect23, 06,080,208,164,42 , 13,106,222,138,21 , 91 , Neo-Grunfeld: Alekhine
595.5,631,521.5,tcect31, 01,064,213,187,35 , 12,103,226,148,11 , 17 , QGA: 3.e4 Nc6 4.Nf3
495.5,621,521.5,tcect13, 15,131,224,108,22 , 11,109,228,130,22 , 29 , English Lukin
618.5,583,515.5,tcect18, 05,049,202,192,52 , 11,114,228,127,20 , 16 , Keres
530.0,652,511.0,tcect40, 13,094,226,154,13 , 08,121,228,127,16 , 08 , Caro-Kann: Advance
574.5,575,529.5,tcect25, 13,070,215,159,43 , 07,113,229,116,35 , 16 , KI: Samisch,
632.0,596,492.0,tcect20, 02,046,193,204,55 , 13,127,230,123,07 , 04 , Spanish: Modern Steinitz
594.5,615,540.5,tcect39, 05,061,214,180,40 , 08,091,232,150,19 , 11 , Dutch
685.0,546,497.0,tcect19, 02,023,162,229,85 , 06,132,233,120,09 , 00 , Trompowsky
633.0,588,506.0,tcect05, 02,045,197,197,59 , 13,113,234,129,11 , 04 , SemiSlav - Moscow
628.0,514,529.0,tcect42, 07,055,193,165,80 , 13,101,234,119,33 , 14 , Sicilian: Taimanov, 6.f4
707.5,429,499.5,tcect15, 06,032,133,199,130, 15,120,235,111,19 , 05 , KID 6.h3
675.0,544,527.0,tcect36, 01,033,167,213,86 , 08,098,236,148,10 , 09 , Scandinavian

542.5,637,524.5,tcect37, 11,086,236,141,26 , 12,096,242,131,19 , 12 , KI, Petrosian, Stein
484.0,636,494.0,tcect26, 15,128,249,090,18 , 20,115,242,103,20 , 01 , Scheveningen, Classical
654.0,592,511.0,tcect38, 04,027,191,213,65 , 04,116,245,124,11 , 03 , Scotch: 4.Nxd4 Bb4+
298.0,456,529.0,tcect48, 123,184,167,26,00 , 16,088,246,122,28 , 13 , KGA: 3.Nf3 g5 4.h4
617.5,591,538.5,tcect45, 02,050,218,171,59 , 10,085,246,136,23 , 11 , Benko Gambit
593.0,638,523.0,tcect46, 01,067,222,165,45 , 04,104,249,128,15 , 06 , Spanish:Chigorin,Keres
732.5,463,554.5,tcect44, 01,015,135,216,133, 04,072,250,159,15 , 07 , French: Winawer
604.0,662,494.0,tcect17, 01,032,253,186,28 , 20,106,251,112,11 , 08 , QGA: 3.e4 Nf6 4.e5
507.0,656,506.0,tcect11, 12,109,252,107,20 , 14,106,252,110,18 , 16 , Nimzo-Indian: Saemisch
628.0,640,502.0,tcect14, 01,037,216,197,49 , 05,118,253,116,08 , 13 , Alekhine
661.5,587,526.5,tcect27, 01,025,198,202,74 , 08,089,255,138,10 , 05 , Slav:  6.Nh4
587.5,673,498.5,tcect08, 05,052,237,175,31 , 07,115,259,112,07 , 18 , CaroCann - Tartakower
616.0,662,505.0,tcect07, 01,036,234,188,41 , 08,106,261,118,07 , 07 , Budapest
622.5,637,489.5,tcect04, 01,034,234,181,50 , 16,109,262,106,07 , 01 , RuyLopez - Smyslow
536.5,707,509.5,tcect01, 05,080,271,125,19 , 09,095,275,110,11 , 21 , none                     
572.5,717,521.5,tcect41, 01,050,274,153,22 , 07,083,280,120,10 , 20 , Bogo-Indian
609.0,718,493.0,tcect12, 01,023,262,185,29 , 04,110,285,098,03 , 18 , Philidor
506.0,740,531.0,tcect21, 11,080,302,100,07 , 08,065,298,115,14 , 26 , Colle: 3...e6 4.Bd3 c5



A: whitescore
B: draws
C: score of asmFish-Nov07 vs. asmFish-Oct03
D: book
E-I : pairs with 0,0.5,1,1.5,2 points for white
J-N : pairs with 0,0.5,1,1.5,2 points for Nov07
O: draws before move 25
P: opening name
Up Topic The Rybka Lounge / Computer Chess / Opening selection for the TCEC 10 superfinal

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill