Toga LOG User Manual
Contents
1. 5 50 Bonus rank 10 100 Bonus rank Candidate scoring builds from a base value 5 10 and a value depending on advancement There is a Bonus function which generates the following values depending on the rank Rank Bonus 0 0 0 1 0 6 2 3 4 5 0 3 6 7 1 For Figure 4 I get the following evaluation b3 isolated 11 10 20 d4 backward and open 14 16 10 Figure 9 Pawns 4 5 Tempo If it s White s move 20 10 is added else 20 10 This factor obviously is there to achieve better comparison of lines with even uneven depth For manually comparing positions in the logfile however it is confusing so to the right of the orange total value I print the total value without this factor 4 6 Pattern A few simple concepts the engine cannot access by searching are implemented here I show them for White for Black they are mirrored p P means Black White Pawn Trapped Bishop 100 100 recognized in one of the following constellations e Ba7 pb6 Bb8 pc7 Bh7 pg6 Bg8 pf7 Ba6 pb5 Bh6 pg5 Blocked Bishop 50 50 e Bel Pd2 piece on d3 e Bfl Pe2 piece on e3 Blocked Rook 50 50 e Ral or Ra2 or Rb1 Kbl or Kcl e Rhl or Rh2 or Rgl Kfl or Kgl You will seldomly find values for this in the log file just because it s avoided by both sides Still the engine needs it in the evaluation to see that it would be bad 4 7 Piec2. 40 30 20 15 15 20 30 40 20 10 0 5 5 0 10 20 30 20 HIG aS gT A 30 5 5 15 20 20 15 Ds 2D 20 10 0 5 5 0 10 20 5 5 15 20 20 15 ari LS S I0 10 57 515 10 0 10 15 15 10 0 10 15 ES 5 10 10 5 5 215 20 10 0 5 5 0 10 20 20 10 0 5 5 0 10 20 35 25 15 10 10 15 gt 25 239 30 20 100 5 5 10 gt 20 30 50 40 30 25 25 30 40 50 40 30 20 15 15 20 30 40 White Bishop Opening White Bishop Endgame 8 6 4 sii 12 07 OP Shoe SLB 8 On 2 0 Q 2 Oo 8 12 6 023 0 0 3 6 12 6 2 4 2 2 4 2 6 9 3 0 3 3 Qu 3 So 4 0 2 8 8 2 0 4 6 0 3 6 6 3 0 6 4 0 2 8 8 2 0 4 6 0 3 6 6 3 0 6 30052 4 2 2 4 2 6 95 23 0 3 3 0 53 29 8 Oo 2 0 OF 52 Oo 8 o k 6 153 0 0 3 6 12 18 18 16 14 14 16 18 18 Br TZ 9 26 560 9 She 18 White Rook Opening White Rook Endgame 0 0 6 3 0 3 3 O 3 6 0 0 0 0 0 0 0 0 SEA 3 0 3 3 0 3 6 0 0 0 0 0 0 0 0 6 3 0 3 3 O 32 56 0 0 0 0 0 0 0 0 E 0 3 k 0 3 6 0 0 0 0 0 0 0 0 a 0 3 3 0 3 56 0 0 0 0 0 0 0 0 26 3 0 3 3 Ov 335 6 0 0 0 0 0 0 0 0 6 lt 3 0 3 3 0 3 6 0 0 0 0 0 0 0 0 White Queen Opening White Queen Endgame 0 0 24 16 12 8 8 12 16 24 0 0 0 0 0 0 0 0 16 8 4 0 0 4 8 16 0 0 0 0 0 0 0 0 12 4 0 4 4 0 4 12 0 0 0 0 0 0 0 0 8 0 4 8 8 4 0 8 0 0 0 0 0 0 0 0 8 0 4 8 8 4 0 8 0 0 0 0 0 0 0 0 12 4 0 4 4 0 4 12 0 0 0 0 0 0 0 0 16 8 4 0 0 4 8 16 655 Hb SO Sor eD Eh 5 24 16 12 8 8 12 16 24 Whi
3. 0 0 0 00 0 0 7 0 0 O0 O O 0 0 0 6 0 0 4 5 5 4 0 0 5 0 2 5 w w 5 2 0 410 2 5 vw w 5 2 0 30 0 0 0 00 0 0 2 0 0 0 0 00 0 0 1 0 0 0 0 0 0 010 A B C D E F G H Scoring KnightOutpostMatrix square If the Knight is protected by 2 Pawns it scores twice In this case I write the e outpost score into the 7 Rank column which is otherwise not used for the Knight 4 7 4 7 Rank Applies for Rook and Queen in one or both of the following conditions e Rook or Queen is on the 7 rank and least one hostile pawn is on the 7 rank e Rook or Queen is on the 7 rank and the hostile King is on th 8 rank Scoring Rook on 7 20 40 Queen on 7 10 20 4 7 5 King Distance Applies for the Queen only and evaluates distance in ranks plus distance in files to the hostile King Scoring Queen s King Distance 10 distance ranks distance files For Figure 4 I get the following evaluation Pieces Bos Piece Mobility OpenLine Outpost King Distance TthRank hs 6 Rel 4 4 8 o 0 0 Ral 7 6 12 10 10 10 Bg2 5 5 5 Ne2 8 8 8 Figure 10 Piece Evaluation 4 8 King This evaluates the concepts of Pawn Shelter Pawn Storm and Piece Attack It only adds to the opening evaluation and is calculated only if the opponent has the Queen and at least one other piece 4 8 1 Pawn Shelter For each of the 3 friendly pawns on the King s file and the 2 adjacent files it calculates
4. The advanced Black f Pawn adds 10 for 4 rank so the result is 32 Black has an advanced f Pawn which gives a penalty of 36 42 20 The King attack by the Black Queen is a bit strange since the Queen can only go to d2 which is 3 squares away from the King But for some reason I m not sure if it s a bug the blocking square counts So it sees the Black Queen can go to e2 it does not consider capturing since it produces the same result with a Black piece on e2 which is 2 squares away from the White King and considers this as an attack Anyway since KingAttackWeight for 1 piece is 0 no attack gets counted Changing the do while into a while in attack cpp line 487 would result in expected behavior The White Queen clearly attacks the Black King according to the definition but the single piece attack also does not count 2 more samples Figure 12 Sample Positions Left Position White Pawns The Pawn in front of the White King is advanced to the 5 rank resulting in 36 32 27 2 for King s file 54 No Black Pawn has advanced White retained castling rights for both directions The value for cl is 36 White Pawn is missing on that file 2 10 Black Pawn has advanced to 4 rank 82 The value for gl is 11 no weaknesses except backrank It takes the average between 54 and 11 32 Black Pawns The d Pawn is missing 36 White s e Pawn has advanced to 5 rank 30 66
5. a penalty based on the advance of the pawn It calculates the distance of the Pawn to the 8 rank and applies the formula 36 distance to 8 as a penalty value For the 2 rank this results in 36 6 0 for the 3 in 36 5 11 etc If there is no Pawn on a row it subtracts a penalty of 36 The penalty for the King s file is doubled If no penalty results from this calculation it gives a penalty of 11 because of the weak backrank Score fpap friendly pawn advance penalty 36 distance to gy Pawn Shelter 2 fpap for the King s file fpap for each of the adjacent files 11 if this results in O 4 8 2 Pawn Storm For this concept it examines the hostile Pawn advances on the 3 files in front of the King For each Pawn on the 4 rank it counts 10 for each on the 5 it counts 30 for each hostile Pawn on the 6 rank it counts 60 Hostile pawns on the 7 rank are not counted All 3 Files have the same weight no doubling here on the King s file Score hpap hostile pawn advance penalty 10 for 4 30 for 5 60 for 6 rank Pawn Storm hpap for the King s file hpap for each of the 2 adjacent files to the left and right If the King retained the right to castle it calculates the same penalty for the King s destination square after castling if castling is possible for both sides it takes the destination square with the smaller penalty and then calculates the average from the King s current squares
6. be disturbing when the viewer loads the logfile for the next depth while you are reading Therefore I prefer using some fixed depth which the engine reaches immediately e g 9 which is the same value I enter into the Toga Min Log Depth engine parameter so the viewer follows the navigation in the Gui smoothly When you press the Load Paste button in this mode you deliberately update to the highest available depth The last feature is top which tells the viewer to stay on top of the Gui Btw when I copied positions back from the viewer into the Gui by selecting the Fen string with the mouse and pressing Ctrl c I sometimes got an Illegal Position error from the Gui I found that I had copied a Tab character before the Fen accidentally which was not accepted I could avoid this by selecting the text dragging the mouse from beginning to end instead as from end to beginning in the viewer The Logviewer requires the free NET 2 0 Runtime 6 Appendix The next two pages show the Position Tables for all pieces White Pawn Opening White Pawn Endgame 15 15 LS 5 0 5 5 0 25 15 0 0 0 0 0 0 0 0 HS 5 0 5 5 P 5 15 0 0 0 0 0 0 0 0 15 e5 O 15 15 0 5 15 0 0 0 0 0 0 0 0 15 5 0 25 25 0 5 15 0 0 0 0 0 0 0 0 15 lt 5 OQ 15 15 0 prag 0 0 0 0 0 0 0 0 AS 0 5 5 0 SSS 0 0 0 0 0 0 0 0 Fb 5 0 5 5 O So 5 0 0 0 0 0 0 0 0 White Knight Opening White Knight Endgame 5135 25 15 10 10 15 525 1359
7. for King on e8 For King on c8 it s 36 missing d Pawn 11 2 advanced c Pawn no Pawn Storm 47 For King on g8 it s 11 Average between 66 and 11 38 After castling kingside both sides would have only 11 so in conjunction with the Position Table values both sides have a strong desire to castle Piece Attacks The only piece attack comes from the White Queen which is blocked at c6 within 2 squares from the Black King The single piece attack does not count King score in Figure 12 left 32 38 6 Right Position White Pawns The Pawn in front of the White King is advanced to the 4 rank resulting in 36 4 20 2 for the King s file 40 The Pawn to the left of the White King is advanced to the gh rank 36 52 11 No black pawn has advanced resulting in 51 for White s Pawn evaluation Black Pawns No Pawn is right to the Black King resulting in 36 White s h Pawn has advanced to rank 4 counting 10 so Black s Pawn evaluation is 46 Piece Attacks The White King is attacked by Qh5 80 Rg7 40 Bd7 20 and Be7 20 but not by Rf8 which cannot reach a square adjacent to the White King The score is 160 0 88 KingAttack Weight for 4 attacking pieces 140 The Black King is attacked by Qc2 but one attacker does not count King score in Figure 12 right 51 46 140 145 Figure 12 right is a position from an interesting game Rybka Toga Toga had
8. in a number from 0 in the opening to 24 in the endgame I log this as a percentage value where O means all pieces are on the board 100 that only Kings and Pawns are left For Figure 2 above a phase value of 16 is displayed meaning it s close to the opening since only a Bishop and a Knight are missing as you see in Figure 4 At the end of the main line in Figure 3 the phase increased to 50 Dor each evaluation block I log the weighted value and then in brackets the opening and endgame values 4 2 Material Queen 975 975 Rook 500 500 Bishop 325 325 Bishop pair 50 50 Knight 325 325 Pawn 70 90 N 325 325 325 Figure 5 Material For the position in Figure 4 we see that White has a Knight opposed to a Bishop of Black Black has the Bishop pair and is a Pawn up 4 3 Position Table For each piece there is a table with 64 entries one for each square Each pieces position is just looked up in the appropriate table and the score is summed up I append the position tables at the end of the document see Appendix 16 01 3 Kg1 33 50 48 Qhs5 1 0 8 Re1 0 0 0 Rd1 2 3 0 Bg2 1 0 6 Ne2 9 10 5 Pieces 24 43 67 f2 0 0 0 h2 12 15 0 b3 4 5 0 g3 4 5 0 d4 20 25 0 es 12 15 0 Pawns 12 15 0 Figure 6 Position Table For Figure 4 we see that both Kings are on the best squares 33 but this will change in the endgame to 48 to get the King
9. penalty and the castling destination squares penalty as the penalty score 4 8 3 Piece Attack A Knight is considered to attack the hostile King if it can reach a field adjacent to the king within one move A Bishop Rook or Queen is considered to attack the hostile King if 1 ona board with only the hostile King and the piece on it it could reach a square adjacent to the King within one move and 2 on the real board without capturing can reach a square lying within 2 squares 2 King moves distance from the King within one move A King attack by a piece weighs by the pieces quality Knight 1 Bishop 1 Rook 2 Queen 4 These values are summed up for all attacking pieces The score for King attack from pieces is then calculated as Piece King Attack 20 sum of attacking values KingAttackWeight number of attackers KingAttackWeight is a function defined as Nr of Attackers Weight 1 0 0 5 0 75 0 88 0 94 0 97 0 99 SY DA On BY bo So a King attack by only one piece does not count at all For Figure 4 I get the following evaluation White Pawn Shelter 22 Black Pawn Storm 10 Black Pawn Shelter 20 White Pawn Storm 0 Attack to white King 80 0 00 for 1 pieces 0 King Attack by Qh5 80 Attack to black King 80 0 00 for 1 pieces 0 Figure 11 King Attack White has an advanced g Pawn resulting in 36 52 2 for the King s file 22
10. to logfiles from the CB Gui If you want to view an existing logfile for a position displayed in the Gui call copy position in the Gui and then press the Load Paste button in the Logviewer to see the logfile If logfiles for different depths are available use the dropdown to switch between them Even more interesting is live analysis while the engine is running In that case you don t have to copy the position If you press the button and the viewer does not find a valid Fen in the clipboard it just displays the latest logfile written by the engine matching the selected depth Rybka 2 2n2 32 bit Toga Il 1 2 1a FastOne Blitz 25 B21 0 Toga Logviewer Datei Bearbeiten Ansicht Einf gen Partie Engine Extras Fenster Hife BS gt NE eaP Position 1rb3rk 3n2qp 8 Q2Npp2 P7 1P6 2P2PP1 3RR1K1 w 11 Depth 9 v auto v top Position 1rb3rk3n2qp 8 Q2Npp2 P7 1P6 2P2PP1 3RRIK1 w 11 15 9293 f5f4 d1d3 c8b7 c2c4 g7h6 d3f3 h6h5 g1g2 d7f6 a5d2 f6d5 c4d5 b8d8 f3d3 f4g3 f2g3 d8f8 d3c3 Current Position P 74 70 30 Kg1 29 50 48 Qa5 1 0 8 Rot 2 3 0 Ret 2 3 0 Nd5 17 20 10 ad 11 15 0 b3 3 5 0 c2 0 0 0 12 0 0 0 92 350 Pieces 50 76 46 Pawns 19 25 0 Pawns 7 Tempo 17610 Notation Partieformular Buch Rybka 2 2n2 32 bit Toga Il 1 2 1a 0 1 B21 FastOne Blitz 25 08 01 2007 f 0 03
11. 0 05 W 14 2 ply 54kNs 4 B 13 8 ply 526kN s 1 e4 c5 2 d4 Beide letzter Buchzug cxd4 0 05 13 33 3 Sf3 0 03 14 39 Sf6 0 06 14 55 4 e5 0 03 15 1 04 Dxd4 Sd5 0 14 14 36 5 Lc4 0 00 15 1 56 c3 Sb6 0 33 13 26 Sc6 6 Le2 0 02 15 59 Lb3 Sc6 0 03 14 29 7 0 0 0 03 14 52 g6 0 07 14 32 8 Sxd4 0 07 15 54 c3 Sxe5 0 14 13 31 9 Sc3 0 05 15 58 a4 Lg7 0 27 14 26 10 a4 0 07 14 30 a5 0 23 15 54 d5 11 h4 0 11 13 58 f4 d5 0 30 13 25 0 0 12 h5 0 11 13 47 040 0 39 13 23 13 Te1 0 24 12 1 05 16 Sc6 0 38 13 45 14 Le3 0 40 12 49 gxh5 0 53 13 1 03 e5 15 Sxc6 0 30 13 28 bxc6 0 59 13 14 16 Ld4 0 37 14 19 Lxd4 0 57 14 32 17 Dxd4 0 44 14 26 Sd7 0 80 14 1 21 Lf5 18 Lxh5 0 36 13 43 e5 0 70 14 24 19 Dd2 0 39 13 24 fS 0 69 13 17 Tb8 20 Tad1 0 45 11 29 Tb8 0 96 14 31 21 b3 0 47 13 26 Kh8 1 09 14 26 Df6 22 Lf3 0 65 12 28 Dh6 Df6 1 55 13 40 Tg8 23 Lxd5 0 78 13 37 Sxd5 cxd5 1 81 12 24 24 Sxd5 0 95 14 12 Dg7 2 10 14 23 25 Dxa5 1 14114 17 MRE 2 0914 20 26 93 1 28 13 12 f4 2 46 14 21 27 Td3 151 14 24 Lb7 2 44 13 21 28 c4 154 14 31 Dh6 244 13 20 Tac8 29 Dc3 148 11 13 Lxd5 2 63 12 22 Tg5 30 cxd5 1 95 13 20 Tbc8 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 36 38 a ETRE EY IEEE Erledigt Standardkiebitz ein ausschalten 1165 8000p Figure 16 Toga Logviewer Still more automatism provides the auto option where the tool loads each logfile automatically after creation It may
12. 1 Preface I m a software developer and occasional chess player and have been fascinated by the Toga source code from the time I was pointed to it from some cc forum I never had considered to spend weeks of my sparetime to create a decent playing engine by myself and add it to the hundreds of them already existing But with Toga we got a readymade real top engine playing well above 2800 Elo and beating most commercial engines amongst them Fritz which has been developed professionally for more than 15 years I expected such an engine code to contain quite some amount of sophisticated chess knowledge to achieve this strength and found that it does indeed Having a strong chess engine is one thing But while these programs play at the level of the strongest human Grandmasters today to the intermediate chess player they present nothing more than one single evaluation number plus a line of moves considered best for both sides the result from an averaging of countless subtle factors You may trust this evaluation value or not latest engine results against human Grandmasters hint that you should trust it but unless you aren t a Master yourself you don t stand any chance to understand a top engine s evaluation of a position So I thought it would be an interesting experience to log every single detail from the engine s evaluation process and make this information available during analysis to provide an average chess player with a bett
13. 5 5 15 4 OF SF SB BR 2 0 4 15 5 35 10100 2S 5 5 15 6 A E 2 6 20 10 O 255 25 0 10 20 8 0 2 0 0 2 0 8 30 20 10 5 5 TOs 20 30 8 8 6 4 4 6 8 8 40 30 20 15 15 20 30 40 Black Rook Opening Black Rook Endgame 0 6 3 0 3 43 0 3 6 0 0 0 0 0 0 0 0 6 3 oe 0 3 6 0 0 0 0 0 0 0 0 6 3 033 3 0 3 6 0 0 0 0 0 0 0 0 6 3 0 43 0 3 6 0 0 0 0 0 0 0 0 6 3 a 3 3 0 3 6 0 0 0 0 0 0 0 0 6 3 0 3 3 0 3 6 0 0 0 0 0 0 0 0 6 3 0 53 3 0 3 6 0 0 0 0 0 0 0 0 Black Queen Opening Black Queen Endgame 5 5 5 5 5 5 5 5 24 16 12 8 8 12 16 24 0 0 0 0 0 0 0 0 16 8 4 0 0 4 8 16 0 0 0 0 0 0 0 0 12 4 0 4 4 0 4 12 0 0 0 0 0 0 0 0 8 0 4 8 8 4 0 8 0 0 0 0 0 0 0 0 8 0 54 38 28 4 0 8 0 0 0 0 0 0 0 0 12 4 0 4 4 0 4 12 0 0 0 0 0 0 0 0 16 8 4 0 0 4 8 16 0 0 0 0 0 0 0 0 24 16 12 8 8 12 16 24 Black King Opening Black King Endgame 40 50 30 10 10 30 50 40 72 48 36 24 24 36 48 72 30 40 20 0 0 20 40 30 48 24 12 0 0 12 24 48 10 20 0 20 20 0 20 10 36 12 o 12 12 0 12 36 0 10 10 30 30 10 10 0 24 0 12 24 24 12 0 24 10 0 20 40 40 20 o 10 24 O 12 24 24 12 0 24 20 10 30 50 50 30 10 20 36 12 A 12 0 12 36 30 20 40 60 60 40 20 30 48 24 12 0 0 12 24 48 40 30 50 70 70 50 30 40 72 48 36 24 24 36 48 72
14. a LOG creates one logfile in HTML format each time it completes one search depth for depth values defined by the new engine parameters Toga Min Log Depth default 11 and Toga Max Log Depth default 15 e The logfiles are stored in the directory C Toga LOG file names are built from the positions Fen and the search depth e You can choose between compact and verbose logging by the new engine parameter Toga Verbose Logging default on e Logging can be used in every engine mode like playing analysis engine match or auto annotation e Logging does not slow down the engine anyway my build seems to be a bit slower than the official release e Logging only works correctly for default engine parameters except the new logging parameters Toga Min Log Depth Toga Max Log Depth and Toga Verbose Logging e A tool Toga Logviewer exists to quickly open the logfile for a certain position 3 Logfile Content The first table within each logfile contains header info position search depth value and main line Position 11 r3rk1 3bb 1pp 2p1p3 1p2Pp1Q 2pP4 1P4P 1 q3NPBP 2RR2K1 w 11 Depth 17 alue Line b3c4 b5c4 e2c3 a2a3 h5e2 b8b4 c3b1 a3a4 b1c3 a4b3 e2e3 b3b2 c1b1 b2c2 b1c1 c2b3 c1b1 f5f4 b1b3 f4e3 Figure 1 Header Table The next table shows the static evaluation of the current position Current Position alue Phase Hina Figure 2 Current Posi
15. e An extensive section accessing the concepts of Mobilty Rook on Open File Knight on Outpost Rook Queen on 7 Rank Queen s King Distance 4 7 1 Mobility This counts the possible moves for each piece without captures Scoring Knight moves 4 4 Bishop moves 6 5 Rook moves 7 2 in the opening 4 in the endgame For the Queen Mobility calculation was implemented the same way originally but than replaced by King Distance evaluation 4 7 2 Open File Applies for Rooks only There is a distinction between a semiopen file no friendly but hostile pawn and a full open file no pawns at all Additional bonus is given if the Rook is on the same semiopen or open file as the hostile King or an adjacent file to the hostile King Scoring Rook on closed file 10 10 Rook on semiopen file 0 0 Rook on semiopen file adjacent to hostile King 10 0 Rook on semiopen file same as hostile King 20 0 Rook on open file 10 10 Rook on open file adjacent to hostile King 20 10 Rook on open file same as hostile King 30 10 4 7 3 Outpost Valid only for Knights in following conditions e the Knight is protected by at least one friendly pawn it s not checked if it can get chased or exchanged e the Knight s square has a positive value in the KnightOutpostMatrix The KnightOutpostMatrix for White for Black it s mirrored is defined as 8 0
16. e is 76 Qe2 Rxg2 77 Nxg2 Bxf2 78 Qxf2 Rxf2 79 Rxf2 where White has 2 Rooks for Queen and Pawn I sometimes read that Rybka plays human like Indeed here it seems that Toga s unexpected strike had put her down and she stopped fighting So in this game I would say Rybka beat herself 4 9 Passed Pawns A pawn is passed if it is open and on the 2 adjacent files to the left and the right of the pawn no hostile pawn is ahead for Be7 it sees the adjacent square for rule 1 at the imaginary position i3 which might be a bug according to the authors intention But within this algorithm nothing is really correct and nothing is really wrong so it doesn t matter much In this position it s actually good to count Be7 as an attacker and it should even count Rf8 Scoring is similar to candidate a passed pawn does not count as a candidate passed status is the next level to promotion so to say For opening score it is Passed Pawn Opening 10 60 Bonus rank For endgame score 3 more concepts are added 1 distance of the Kings in moves to the field in front of the pawn evaluation is 20 distance of hostile King 5 distance of friendly King 2 a passed Pawn is considered free if the square in front of it is empty and it can securely advance one step without getting lost score 60 3 apassed Pawn is unstoppable if the opponent has no pieces just King and Pawns and a no friendly piece is in the way to queening and b t
17. er understanding of chess positions Since the logfiles still contain only numbers albeit in great detail and layouted nicely in concise HTML tables a user needs some background information to be able to read and understand them I provide this knowledge throughout the oncoming 20 pages After studying this paper you ll know erverything about the evaluation of this strong engine I ll explain it in natural language without using any programming terms and can manually evaluate a position in the same way Toga does it within one minute But you won t be able to repeat the evaluation 500000 times a second while trying out different lines My solution actually does not fit very well for explaining tactics For accessing this area you still have to play around while the engine is running and observe how the evaluation changes But once you found a line with a high or low evaluation score the logfile does explain where this value comes from I read that a commercial visualization tool for the Rybka engine will become available around summer see www uciengines de UCI Engines Rybka vriv vriv html and this is exactly what I m longing for to get from this tool an explanation about engine tatics along with the positional analysis With the term Toga I refer to the open source chess engine created as Fruit 2 1 by Fabien Letouzey and enhanced as Toga II by Thomas Gaksch 2 Toga Logfile Creation e The engine version Tog
18. he hostile King cannot reach the Pawn before queening score 800 4 a passed Pawn is unstoppable also if the opponent has no pieces just King and Pawns and a the friendly King is on an adjacent square to the queening square and b the friendly King is on an adjacent square to the Pawn score 800 Figure 13 Sample Positions If in Figure 13 left if it s White s move a4 is unstoppable g5 is free If it s Black s move g5 is unstoppable a4 is free rules 2 and 3 In the right position e6 is unstoppable in any case rule 4 Endgame score is Passed Pawns Endgame 20 120 Kings distances score free score unstoppable score Bonus rank I show the Passed Pawns evaluation in one more sample position Figure 14 Sample Position Opening 10 60 0 3 for rank 5 28 Endgame 20 120 80 20 distance white King 25 5 distance black King 0 3 for rank 5 73 Figure 15 Passed Pawns In Figure 14 the White King is 4 moves away from the square c3 in front of the Passed Pawn at c4 20 4 80 while the Black King is 5 moves away from that square 5 5 25 4 10 Draw Evaluation Toga implements draw evaluation for move repetition and the 50 moves rule which may overwrite the evaluation value with 0 I do not log this If in a logfile the total is 0 and does not match with the subvalues the engine found a draw 5 Toga Logviewer The Toga Logviewer is a tool for quick access
19. into the center Most valuable for White are the Pawns on d4 20 and eS 12 for Black e6 12 all positive scores for Black result in negative values worst evaluated are the Pawns on h2 and h7 12 for the whole a and h file that s for getting the engine capturing towards the center 4 4 Pawns For technical reasons Pawn evaluation in Toga is split into 2 blocks Pawns and Passed Pawns which I keep in the logfile The Pawns block implements the concepts of doubled isolated backward open and candidate Pawns 4 4 1 Doubled Only the more advanced pawn counts as doubled In case of a triple pawn the 2 foremost pawns count as doubled 4 4 2 Isolated No friendly pawn exists on the files to the left and right of the pawn 4 4 3 Backward The pawn is not isolated but all pawns on the files to the left and right of the pawn are more advanced than the pawn Even if this is true the pawn still is not considered backward if one of the following rules applies 1 a friendly pawn on a one of the adjacent files to the left and right is more advanced by only one rank and 1 1 on the field in front of the pawn is no friendly or hostile pawn and 1 2 there is no hostile pawn on the file to the left or right 1 or 2 ranks ahead 2 the pawn is still on the second rank a friendly pawn on one of the adjacent files to the left and right is 2 ranks ahead and 2 1 on the fields 1 and 2 ranks ahead of the pawn is no friendly o
20. r hostile pawn and 2 2 there is no hostile pawn on the file to the left or right 1 2 or 3 ranks ahead Some examples Figure 7 Backward Pawns In Pos 1 b4 is backward In Pos 2 b4 is not backward because of rule 1 In Pos 3 and Pos 4 b4 is backward because of rule 1 1 In Pos 5 and Pos 6 b4 is backward because of rule 1 2 In Pos 7 b4 is not backward In Pos 8 b2 is not backward because of rule 2 1 In Pos 9 b2 is backward because of rule 2 1 In Pos 10 b2 is backward because of rule 2 2 4 4 4 Open There is no friendly or hostile pawn in front of the pawn 4 4 5 Candidate A pawn is considered a candidate if it is open and both of the following rules apply 1 the number of hostile pawns on the adjacent files to the left and right in front of the pawn is not bigger than the number of friendly pawns on the adjacent files left and right behind or on equal row to the pawn 2 the number of hostile pawns currently attacking the pawn is not bigger than the number of friendly pawns defending the pawn Some examples Pos 3 Figure 8 Candidate Pawns In Pos 1 b4 is no candidate because it s not open In Pos 2 b4 is a candidate In Pos 3 b4 is no candidate because of rule 1 In Pos 4 b6 is no candidate because of rule 2 but can get one by advancing one step In Pos 5 b5 is a candidate 4 4 6 Scoring Doubled 10 20 Isolated 10 20 Isolated and Open 20 20 Backward 8 10 Backward and Open 16 10 Candidate
21. sacrificed 2 pawns for a strong King attack early in the game but at some point couldn t make progress Rybka failed to find a way to simplify either Both sides had been shifting pieces forth and back for more then 50 moves the game looked completely drawish Static evaluation by Toga is 71 after the expected move 74 Qe4 even 93 At a depth of 17 Rybka evaluates the position with 72 Toga sees 36 But suddenly in the same position she had been holding accurately for such a long time Rybka played the tactical blunder 74 Re1 after which Toga finally got it s attack going with 74 Bxh4 75 Kg1 Rxg3 76 Qd1 Bg4 77 Qd3 Rgf3 78 d6 Bxf2 79 Rxf2 Rxf2 Both engines take exactly 4 s to find Bxh4 after Rel but in the move before Rybka takes 18 s to abandon Rel although it sees Bxh4 already after 11 s Toga just never considers to play Rel You normally would expect that engines after having played around in a constant position for some time have the positions stored in hash memory and do not blunder even when getting low on time The point here was that Rybka recently had played one single pawn move h2 h4 obviously to bypass the 50 move rule since she felt for winning thereby 1 providing Toga an important attack mark 2 rendering her cash useless in a situation where 3 she was running low on time Analysis shows that even after 75 Rxg3 it is not necessary for White to give up the exchange According to Toga best lin
22. te King Opening White King Endgame 40 30 50 70 70 50 30 40 72 48 36 24 24 36 48 72 30 20 40 60 60 40 20 30 48 24 12 0 0 12 24 48 20 10 30 50 50 30 10 20 36 1 2 0 12 12 0 12 36 10 0 20 40 40 20 o 10 24 0 12 24 24 12 0 24 0 10 10 30 30 10 10 0 24 0 12 24 24 12 0 24 10 20 0 20 20 0 20 10 364512 0 AA 12 0 12 36 30 40 20 0 0 20 40 30 48 24 12 0 0 12 24 48 40 50 30 10 10 30 50 40 72 48 36 24 24 36 48 72 Black Pawn Opening Black Pawn Endgame 15 5 a 23 58 0 AR sl 0 0 0 0 0 0 0 0 15 5 OC 5 45 0 D ihe 0 0 0 0 0 0 0 0 15 5 0 15 15 0 some 0 0 0 0 0 0 0 0 15 5 0 25 25 0 5 S 0 0 0 0 0 0 0 0 15 5 0 15 15 0 Be 15 0 0 0 0 0 0 0 0 15 5 0 00 45 0 E Es 0 0 0 0 0 0 0 0 15 5 0 5 5 0 5 15 0 0 0 0 0 0 0 0 15 5 0 D 55 0 E ES 0 0 0 0 0 0 0 0 Black Knight Opening Black Knight Endgame 50 40 30 25 25 30 40 40 30 20 15 20 30 40 35 25 15 10 10 15 25 35 30 20 10 5 S 10 20 30 20 10 000 5 0 10 20 20 10 0 5 5 0 10 20 10 O 10 15 15 10 o 10 15 5 2563102101 25 5 15 o Tb 1552012018 5 15 B gt 5 10 S10 5 5 15 5 5 315 20 20 18 5 5 20 10 O 25 5 0 10 20 20 10 0 a EY 0 10 20 30 20 10 5 5 10 20 30 135 25 125 20 10 15 25 135 40 30 20 15 15 20 30 40 Black Bishop Opening Black Bishop Endgame 18 18 16 14 14 40 30 20 15 20 30 40 8 0 2 0 0 2 0 8 30 20 10 5 5 10 20 30 6 2 mA 2 2 gt 4 2 6 20 10 0 5 025 0 10 20 4 Oh R238 8 EZ 0 4 15 Bs 55 10 10 1
23. tion The evaluation value is different to the first table because this is a static evaluation while the final value was found after a search of 17 semimoves But it s also interesting what the engine thinks of a position without searching The table in Figure 2 was generated in compact logging mode You can see that the evaluation value orange sums up from 8 building blocks green A negative value always means an advantage for Black a positive value an advantage for White Then logging goes down the main line and prints the evaluation after each move until in the last table it gets to the final evaluation value after 20 semimoves in this case mainline length does not always reflect the search depth Each table displays the moves played so far and the current position Figure 3 End of the Main Line 4 Verbose Logging This section explains a verbose logfile based on the following position Figure 4 Sample Position 4 1 Game Phase Evaluation of certain concepts changes for different phases of a chess game while it evolves from opening to middlegame and endgame Toga has the simple concept of calculating 2 values in each evaluation block one for the opening and one for the endgame It calculates the game phase based on the number of pieces on the board and then takes the weighted average between opening and endgame value Phase 24 4 nr of Queens 2 nr of Rooks nr of Bishops nr of Knights This results
Download Pdf Manuals
Related Search