Tuesday, January 18, 2022

Centipawn analysis in ChessBase

     For years I have been using Fritz 12 to analyze games, first at about 15 seconds per move then going back through the game at a leisurely pace to fine tune the quick analysis. 
     Although Fritz 12 is still the program I use for most analysis, I do have ChessBase 16 on my laptop. It offers “Tactical Analysis" which checks for mistakes and tactical inaccuracies, but then it has something intriguing called "Centipawn Analysis" which has the short explanation that it "analyzes games for precision play in comparison to engine moves (and detect cheating)."
     This analysis uses a new-to-me unit of measure called Average Centipawn Loss. A Centipawn is 1/100th of a Pawn and everyone is familiar with seeing the standard engine evaluation given in that unit of measure, e.g. 0.67 or -1.28, etc. The Centipawn Loss is the difference between the move actually played compared to strongest engine move. I found this chart that gives a rough estimate of average Centipawn losses by class: 
 

How the Centipawn loss is calculated? 
1) The engine measures the value of the position 
2) When a player moves, the value of the new position is compared to the value of the old position. The difference in Centipawns signifies the accuracy of the move.
 
    Note that the value won't go up because you aren't going to play a move that's better than the one the engine thinks is best...at least in the engine's opinion and it's the one doing the figuring. For example, if, after your move the value of the position drops from 0.75 to 0.25 then the loss would be 0.50. 
 
     The idea is that the average Centipawn loss score over the course of the game (or better yet, several games) will give some indication of a players level of skill. As you can see from the chart, Grandmasters will score between a 10 and 20 Centipawn loss for a game while average players will be between 80-100. 
     The lower the average Centipawn loss a player scores in a game, the more accurate his moves were compared to the engine used. If the best engine move is always played, the Centipawn loss for a game is zero. 
     What's the purpose of this evaluation? Unless you're playing Magnus Carlsen or Hikaru Nakamura online and think you're opponent was using an engine you can run the game through a Centipawn analysis to see if it returns a low score. If it does, that may be an indication that an engine was being used. Of course, by that time it would be too late. 
     As a quick test I ran a Centipawn Analysis on the Blumin-Santasire game from the January 12th post. The results were: 
 
Blumin centipawn loss = 67 
Santasiere centipawn loss = 17 
One would hardly call Blumin and average player and the high score suggests that he made at least one serious mistake in what was a rather short game. 
 
     I also ran one of my games, a 26 move draw against an ICCF International Master, played on Lechenicher SchachServer through the Centipawn loss analysis. This was a good test because engine use is allowed on the site. The game was played in 2017 and I am not sure which version of Stockfish I was using...9 perhaps and, of course, I have no way of knowing what engine my opponent was using, but it was probably also Stockfish 9 as it was the strongest engine available. Based on the scores it's pretty clear engines were used! My Centipawn loss = 3 Opponent's Centipawn loss = 6 
     Finally, I tested the following online G10 using the Centipawn Analysis. Naturally, you would expect a high score for both players which is exactly what happened. I don't know what this has to do with anything, but I thought it was interesting. Has anybody found a practical use for the Centipawn Loss function?!
 

Guest - Tartajubow

Result: 0-1

Site: Online G10

Date: 2022

Pickering/Goldsmith Defense

[...] 1.d4
1.e4 h5 2.d4 d6 3.♘c3 h4 4.h3 a6 5.a4 b6 6.♗e3 ♗b7 7.♗d3 ♘d7 8.♘f3 e6 9.♕d2 ♗e7 10.O-O ♘gf6 W. Vachier-LaGrave vs. Wesley So, Lichess G10, 2019. White is much better, but he went on to lose.
1...h5 2.e4 h4 Obviously this is a a very weak response and it's even worse that the Corn Stalk Defense (1...a5) because it does not open any lines for development and it allow white quick and easy development. Additionally weakens the K's position slightly. It's sole merit is that it sometimes leaves white bewildered. 3.♘f3
3.♘c3 c6 4.h3 d5 5.♘f3 dxe4 6.♘xe4 ♗f5 7.♗d3 ♘f6 8.♘xf6+ gxf6 9.c3 If 9.Bxf5 Qa5+ Zorko,J (2454)-Dzhumagaliev,Y 2016. White is better.
3.h3 Might be a little better. 3...c5 4.d5 (4.dxc5 ♕a5+ 5.♘c3 ♕xc5 6.♘f3 ♘c6 is also good for white.) 4...d6 5.♘c3 e5 leaves white well off.
3...h3
3...c6 4.♗g5 h3 5.gxh3 d5 6.exd5 cxd5 7.h4 ♘f6 8.♗xf6 gxf6 Black as equalized. Cherny,O (2046)-Omariev,M (2335)/Moscow 2011
4.♘c3 (4.gxh3 d6 5.♘c3 ♗xh3 6.♗xh3 ♖xh3 Clearly white is better.) 4...hxg2 5.♗xg2 c6 6.♗f4 ♕b6 White need not waste time defending the b-Pawn because black is way behind in development and so taking the P would make his position even worse. 7.♖b1
7.♕d2 ♕xb2 8.♖b1 ♕a3 9.O-O with a huge lead in development plus the Q is badly placed on a3.
7...♘f6 Somewhat better was 7...d6 8.O-O Also good was 8.e5 and white is better after either 8...Nd5 or 8...Nh5. 8...d6 9.♖e1 Now I should have played 9...Nh5 after which white would remain much better, but at least I would have been better off than after the move I played. 9...♗h3 Fritz make the comment, "Black crumbles in a dire situation." and the evaluation soars to over 6 Ps in white's favor. 10.♗xh3 ♖xh3 11.♔h1 Pointless because he has no real prospects on the g-file, but he still retains the advantage.
11.e5 and white has triumphed according to Fritz. 11...♘h5
11...dxe5 12.dxe5 ♘d5 13.♘xd5 cxd5 14.e6 fxe6 15.♕d3 ♘c6 16.♕g6+ ♔d8 17.♘g5 wins.
12.♗d2 dxe5 13.dxe5 f5 14.♔g2 ♖xf3 15.♕xf3 g6 Black's position is horrible.
11...e6 12.♖g1 12.e5 was still playable and the best move. 12...♘bd7 13.♖g3 After this black has fully equalized and it's a whole new game! 13.Ng5 was better, but even then black's position has greatly improved to bear equality. 13...♖xg3 14.fxg3 ♗e7 15.♔g2 O-O-O 16.a4 Hoping for a Q-side attack. 16...♖h8 Hoping for a K-side attack. Actually, a center attack with 16...e5 was somewhat better. 17.b4 As planned. 17...♘g4 As planned except the center punch 17...e5 was still better. 18.a5 ♕c7 19.b5 Not the best because black can still get an equal game by 19...e5
19.♕e2 leads to some interesting play. 19...e5 20.dxe5 ♘dxe5 21.h3 ♘xf3 22.♕xf3 ♘e5 23.♗xe5 dxe5 24.♕xf7 ♖d8 White's advantage is minimal. Note that the g-Pawn is safe because after 25.♕xg7 ♖d2+ 26.♔h1 ♕d7 27.♕h8+ ♗d8 Black is threatening a 2nd rank invasion after ...Qf7 28.♕h5 ♕d4 29.♕f3 and as dangerous as white's g- and h-Pawn look, they are not far enough advanced to be a real threat at this time. 29...♔b8 30.h4 ♗e7 31.b5 ♖f2 32.♕d3 ♖d2 33.♕xd4 exd4 34.♘a4 cxb5 35.♖xb5 ♖xc2 36.♖d5 ♖c4 37.♘b2 ♖c1+ 38.♔g2 ♖c2+ 39.♔f3 ♖xb2 40.h5 ♖h2 41.g4 ♖h3+ 42.♔g2 ♖h4 43.♔g3 ♔c7 44.e5 ♖h1 45.♔f4 ♖f1+ 46.♔e4 ♖e1+ 47.♔f5 (47.♔xd4 loses to 47...♔c6) 47...♖d1 48.h6 ♔c6 49.♔e6 ♖h1 50.♖xd4 ♗c5 51.♖d3 ♖xh6+
19...e5 Finally! 20.♗d2? White has two good choices here. Either 20.dxe5 or 20.dxc5 with equal chances. 20...exd4 21.♘e2 Obviously white intended 21.Nxd4, but realized that it would allow 21...Rh2+ so the threatens to take with the other N.
21.bxc6 dxc3 22.cxd7+ ♕xd7 Also favors black. The c-Pawn is immune because of the fork ...Ne3+
21...c5 22.♘f4 ♘df6 23.b6 Hoping for some Q-side play, but he doesn't get any. (⌓ 23.h3 ♘e5 24.a6−⁠+) 23...axb6 24.axb6 ♕c6 25.♖a1 ♔d7 26.♖b1 ♘xe4 The game is over, but with only a handful of seconds left white makes a gross blunder. 27.♗e1 This move cost a gob of Centipawn points!
27.h3 ♘xd2 28.hxg4
28.♕xd2 ♘e3+ 29.♔f2 g5 30.♘g2 g4 31.hxg4 ♘xg4+ 32.♔e2 ♕e4+ is a slaughter.
28...♘xb1 is winning for black.
27...♘e3+ White resigned. Centipawn losses: White 91, Black 36.
Powered by Aquarium

No comments:

Post a Comment