Results 1 to 6 of 6

Thread: Chess.com engine evaluations

  1. #1
    Join Date
    Jan 2009
    Location
    Tecumseh, ON
    Posts
    3,251
    Blog Entries
    1

    Default Chess.com engine evaluations

    I am often a Luddite when it comes to chess engine evaluations of positions from games that I have played but the engine evaluation on Chess.com is right there when you look at your archive games so occasionally I have been using the feature. I am reminded of why I don't use engines as much as I should when I evaluate the engine suggestions. The engine suggests that a perfectly reasonable move in an advantageous position has swung the evaluation of the position from +3.68 to something like +2.7. I play a logical followup move and we're back to something like +3.88. In the game, I was up a solid pawn in an ending with two rooks, knight, four pawns against two rooks, bishop, three pawns all on the kingside. My opponent had walked himself into an unbreakable pin and was in time trouble to boot. Following the directions of countless grandmasters I improved my position and postponed doing anything committal until all my pieces were on their ideal squares (which meant the king which was the only piece which was not somewhat ideally placed) and my opponents time situation was a bit more critical. I made several somewhat non-commital moves even allowing two-fold repetition before triangulating my king so that we got the same position two more times but with different players on the move. The only way to break the position open was with an exchange sacrifice that would leave him hopelessly lost and still pinned. We traded a pair of rooks to prevent me from just grabbing the pinned bishop and then we had a zugswang with no legal moves for my opponent that didn't immediately lose material and with the opportunity for me to get a king and pawn ending with an e5-pawn and g-h pawns versus just g-h pawns for my opponent at any time I wanted unless he simply gave me a piece which would leave him with a rook versus knight, pawn and rook in his best scenario. He resigned with 2 seconds left on his clock in that hopeless position.

    Out of curiosity, I have previously run positions and games through Komodo and Fritz and Stockfish on my other computer and they do not exhibit the same evaluation swings as the chess.com app does.

  2. #2
    Join Date
    Jan 2009
    Location
    Tecumseh, ON
    Posts
    3,251
    Blog Entries
    1

    Default

    Here is an example of the chess.com engine evaluations throwing shade on a reasonably well-played game. I can't take credit though, I saw the line some 45 years ago in the book "How to Open a Chess Game" and could not believe it when I got to play the line on the board though I think I had an almost identical blitz game once, some 30 or 40 years ago. The original game was annotated by Portisch or Hort. My memory of the details is fading.



    [Event "Live Chess"]
    [Site "Chess.com"]
    [Date "2019.02.08"]
    [Round "?"]
    [White "NN"]
    [Black "vlad drkulec"]
    [Result "0-1"]
    [TimeControl "1800"]
    [WhiteElo "1581"]
    [BlackElo "1785"]
    [Termination "vlad drkulec won by resignation"]
    [ECO ""]
    [CurrentPosition "r2k4/ppp2B2/3p1n2/2b1p3/4P3/2NP1bpr/PPPQ1P2/R4RK1 w - - 0 17"]


    1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O { C50: Giuoco Piano Game: 4.O-O } 4...d6 5.d3 Nf6 6.Bg5 h6 7.Bh4 g5 8.Bg3 h5 { (+0.29 → +2.98) Mistake. The best move was 8... Na5. } ( 8...Na5 9.Bb5+ ) 9.Nxg5 { (+1.10 → -0.62) Mistake. The best move was 9. h4. } ( 9.h4 9...Na5 10.Nbd2 10...g4 11.Ng5 11...Nxc4 12.Nxc4 12...b5 13.Ne3 13...O-O ) 9...h4 10.Nxf7 hxg3 { (-0.74 → +1.82) Blunder. The best move was 10... Qe7. } ( 10...Qe7 11.Nxh8 ) 11.Nxd8 Bg4 12.Qd2 Nd4 13.Nc3 { (+0.50 → ♚ Mate in 7) Blunder. The best move was 13. Ne6. } ( 13.Ne6 13...Ne2+ 14.Qxe2 14...Bxe2 15.Nxc7+ 15...Kd7 16.Nxa8 16...Bxf1 17.Bb5+ 17...Ke7 ) 13...Nf3+ 14.gxf3 Bxf3 15.h3 Rxh3 16.Bf7+ Kxd8 0-1

  3. #3
    Join Date
    Mar 2013
    Posts
    113

    Default

    One possibility is that the chess.com engine is either too weak or too slow to provide accurate evals. For instance, in the game you posted, 13.Nc3 is not a blunder (White is already lost), but after 13.Ne6 Ne2+ IS a blunder. Instead, Black should play 13...Nf3, which forces mate.

    Another possibility for the wildly swinging evals on chess.com is that they may be using the Komodo MonteCarlo engine, which gives evals based on its estimate of the probability of winning given many different possible continuations, rather than on the results of best play from both sides (as the traditional AB engines like Stockfish do).

    BTW, the line you played to win was analyzed in Tartakower's book 500 Master Games, or you might have previously seen the game below, which is the earliest one in ChessBase:

    [Event "St Petersburg"]
    [Site "St Petersburg"]
    [Date "1874.??.??"]
    [Round "?"]
    [White "Knorre, Victor"]
    [Black "Chigorin, Mikhail Ivanovich"]
    [Result "0-1"]
    [ECO "C50"]
    [PlyCount "28"]
    [EventDate "1874.??.??"]
    [EventType "game"]
    [EventRounds "1"]
    [EventCountry "RUS"]
    [SourceTitle "EXT 2011"]
    [Source "ChessBase"]
    [SourceDate "2010.11.26"]
    [SourceVersion "1"]
    [SourceVersionDate "2010.11.26"]
    [SourceQuality "1"]

    1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. O-O Nf6 5. d3 d6 6. Bg5 h6 7. Bh4 g5 8. Bg3
    h5 9. Nxg5 h4 10. Nxf7 hxg3 11. Nxd8 Bg4 12. Qd2 Nd4 13. Nc3 Nf3+ 14. gxf3 Bxf3
    0-1

  4. #4
    Join Date
    Sep 2008
    Location
    Richmond Hill, Ontario
    Posts
    51

    Default

    On my local machine, I run Stockfish-10 ($0 and #1 or close to it) with 2.4GB of memory dedicated to its hash tables (space for working notes so it doesn't have to re-calculate/evaluate positions) and 3 cpu cores.

    On chess.com servers shared by 100's or 1000's of users, I'm sure they cannot devote anywhere near that many resources (mem+cpu) to each user.

    On my front-end GUI (Fritz, with "infinite analysis" turned on), I can set it to show Stockfish-10's current top-N (in my case, I set N=6) moves as it is calculating. Often I see that what it initially had as its #1 move would eventually (10-15 secs, say) move down and sometimes out of the top-6 list. So, the more time/memory/cpu the better it gets at evaluating moves/positions. If I cut off analysis at, say, 1 or 2 seconds then I might also see evaluations swinging from move to move on my local machine.
    Last edited by Don Parakin; 02-09-2019 at 03:40 PM.

  5. #5
    Join Date
    Jan 2009
    Location
    Tecumseh, ON
    Posts
    3,251
    Blog Entries
    1

    Default

    Quote Originally Posted by John Upper View Post
    One possibility is that the chess.com engine is either too weak or too slow to provide accurate evals. For instance, in the game you posted, 13.Nc3 is not a blunder (White is already lost), but after 13.Ne6 Ne2+ IS a blunder. Instead, Black should play 13...Nf3, which forces mate.

    Another possibility for the wildly swinging evals on chess.com is that they may be using the Komodo MonteCarlo engine, which gives evals based on its estimate of the probability of winning given many different possible continuations, rather than on the results of best play from both sides (as the traditional AB engines like Stockfish do).

    BTW, the line you played to win was analyzed in Tartakower's book 500 Master Games, or you might have previously seen the game below, which is the earliest one in ChessBase:

    [Event "St Petersburg"]
    [Site "St Petersburg"]
    [Date "1874.??.??"]
    [Round "?"]
    [White "Knorre, Victor"]
    [Black "Chigorin, Mikhail Ivanovich"]
    [Result "0-1"]
    [ECO "C50"]
    [PlyCount "28"]
    [EventDate "1874.??.??"]
    [EventType "game"]
    [EventRounds "1"]
    [EventCountry "RUS"]
    [SourceTitle "EXT 2011"]
    [Source "ChessBase"]
    [SourceDate "2010.11.26"]
    [SourceVersion "1"]
    [SourceVersionDate "2010.11.26"]
    [SourceQuality "1"]

    1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. O-O Nf6 5. d3 d6 6. Bg5 h6 7. Bh4 g5 8. Bg3
    h5 9. Nxg5 h4 10. Nxf7 hxg3 11. Nxd8 Bg4 12. Qd2 Nd4 13. Nc3 Nf3+ 14. gxf3 Bxf3
    0-1
    I did not remember this game from Tartakower's book though I suspect it was the original example. Looking through "How to Open a Chess Game" that it was actually the chapter by Tigran Petrosian that suggested this line for black as a caution against white's pin with 6.Bg5. The book came out in 1974 so it was 45 years ago that I saw this game. Petrosian does not give the names of the players.

  6. #6
    Join Date
    Mar 2013
    Posts
    113

    Default

    Quote Originally Posted by Vladimir Drkulec View Post
    I did not remember this game from Tartakower's book though I suspect it was the original example. Looking through "How to Open a Chess Game" that it was actually the chapter by Tigran Petrosian that suggested this line for black as a caution against white's pin with 6.Bg5. The book came out in 1974 so it was 45 years ago that I saw this game. Petrosian does not give the names of the players.

    I had seen this line recently in Reti's Masters of the Chessboard, where it appears as a note. It's a really good book, and -- despite the terrible number of typos in the Russell edition -- worth more than the $12 it was on sale for at Strategy Games last year.

    Here are Reti's notes, with analytical corrections by SF9 and a typo or two fixed by me:

    [Event "British CA-05 Casual Games"]
    [Site "London"]
    [Date "1862.06.??"]
    [Round "10"]
    [White "Dubois, Serafino"]
    [Black "Steinitz, William"]
    [Result "0-1"]
    [ECO "C50"]
    [Annotator "Richard Reti +"]
    [PlyCount "74"]
    [EventDate "1862.06.??"]
    [EventType "tourn"]
    [EventCountry "ENG"]
    [SourceTitle "Masters of the Chessboard"]
    [Source "Russell"]
    [SourceDate "2018.10.20"]
    [SourceVersion "1"]
    [SourceVersionDate "2018.10.20"]
    [SourceQuality "1"]

    1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. O-O {According to the modern theory of
    openings, it is better not to castle so early in view of the possibility of
    castling later on the queenside. It must be noted in this connection however,
    that we cannot find this postponement of castling in the games before
    Steinitz, for example, in Morphy's games. Since Morphy always played for a
    rapid advance of his center pawns in order to open the game, he was of course
    bent above all upon protecting his king by means of castling.} Nf6 5. d3 d6 {
    MCB has Black's 4th and 5th moves reversed, so Reti's comment to 5...Nf6 ---
    "In contrast to his opponent, Steinitz quite correctly defers castling" ---
    makes no sense, since Black could not castle before playing ...Nf6.} 6. Bg5 {
    If White could have foreseen the consequences of this move, he would probably
    have moved to e3 instead.} h6 7. Bh4 g5 {We see here the consequences of
    White's premature castling and the postponement of the same move by Black. We
    would vainly look for such an attack which today is accepted as a matter of
    course, in an offensive player of Morphy's type. This is because such an
    advance is appropriate only when the center is fully protected and if possible
    closed up, while Morphy, as we have seen, always endeavored to open the game
    in the center as quickly as possible. In the layout of the attack this game,
    therefore, is a predecessor of Steinitz's later offensive games, being
    characterized by a protected and closed center and by the pawn attack on the
    kingside.} 8. Bg3 h5 9. h4 ({A very fine combination after} 9. Nxg5 {would
    have been} h4 $1 10. Nxf7 hxg3 11. Nxd8 Bg4 12. Qd2 Nd4 13. Nc3 Nf3+ 14. gxf3
    Bxf3 {and an early mate.}) ({SF9 finds improvements here:} 9. Nxg5 h4 10. Nxf7 Qe7 $142 $1 $17 (10...
    hxg3 $6 11. Nxd8 Bg4 $8 12. Qd2 Nd4 13. h3 $8 (13. Nc3 Nf3+ $3 $19 {#6}) 13...
    Ne2+ $8 14. Qxe2 Bxe2 15. Ne6 $1 Bb6 16. Nc3 $1 Bxf1 17. Kxf1 $15) 11. Nxh8
    hxg3 $17) 9... Bg4 10. c3 Qd7 11. d4 exd4 12. e5 dxe5 13. Bxe5 Nxe5 14. Nxe5
    Qf5 $1 {The counterattack of White in the center is thus parried and Black's
    attack intensified.} (14... Bxd1 15. Nxd7 Nxd7 16. Rxd1 O-O-O $17) 15. Nxg4 {
    ?? No comment in MCB, but this goes from bad to lost.} (15. Bxf7+ $142 Kf8 16.
    Qe1 $17) 15... hxg4 16. Bd3 Qd5 17. b4 {This move is made with the intention
    of continuing, after 17... Bb6, with 18.c4 and 19.c5. Black, however, prefer
    to give up the threatened bishop, presuming that the fate of the attack will
    be decided on the h-file. The point of the attack is to be found in move 22 of
    Black, Steinitz exchanging queens in spite of the fact that he has one piece
    less, order to prevent the escape of the white king to f1.} O-O-O 18. c4 {
    The best move under the circumstances. It is intended to prevent the black
    queen from dominating the f5-square or the long diagonal, in order to be able
    to play g3 if necessary. The importance of the f5-square will become apparent
    in the continuation of the game.} ({Not in MCB:} 18. bxc5 $4 Rxh4 $19) 18...
    Qc6 19. bxc5 Rxh4 20. f3 Rdh8 21. fxg4 Qe8 {?? Not in MCB.} (21... Nxg4 $142
    $19 22. Bf5+ Kb8 23. Bxg4 d3 $8) (21... Ne4 $142 $19) (21... Kb8 $142 $19) 22.
    Qe2 (22. Re1 Rh1+ 23. Kf2 Qxe1+ 24. Qxe1 Nxg4+ 25. Ke2 Re8+ $19) ({Not in MCB:
    } 22. Bf5+ $8 $11 {this check defends g4 and so prevents the tactics with ...
    Nxg4+ that Reti mentioned in his note.} Kb8 23. Re1 $8 $11) 22... Qe3+ $1 23.
    Qxe3 dxe3 24. g3 Rh1+ 25. Kg2 R8h2+ 26. Kf3 Rxf1+ 27. Bxf1 Rf2+ 28. Kxe3 Rxf1 {
    On account of the blocked up position of his pieces, it is evident that the
    endgame is hopeless for White. Black won after a few more moves.} 29. a4 Kd7
    30. Kd3 Nxg4 31. Kc3 Ne3 32. Ra2 Rxb1 33. Rd2+ Kc6 34. Re2 Rc1+ 35. Kd2 Rc2+
    36. Kxe3 Rxe2+ 37. Kxe2 f5 {White resigns} 0-1

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •