|
< Earlier Kibitzing · PAGE 3 OF 3 ·
Later Kibitzing> |
Jun-13-11
 | | kutztown46: Part 8:
After 55...Kc6, <AK> was hoping that Rybka would play 56. Bd6. Indeed, at low ply during the game, I was worried because Stockfish "saw" that move, and temporarily assigned it an eval of over 1.00. However, Stockfish gives the position this strange evaluation: After 55...Kc6:
click for larger viewAnalysis by Stockfish 2.0.1 JA 64bit (31-ply):
1. ² (0.60): 56.Kf4 Bc4 57.Bd4 Be7 58.Kf3 Kd7 59.Be3 Rf8 60.Nf4 Rfg8 61.Bb6 Kc6 62.Bd4 Kd7 63.Re1 Rh7 64.Bb6 Rhh8 65.Ba5 Rh7 66.Rg1 Bc5 67.Rh1 Be7 68.Rdh2 Bb3 69.Bb4 Bd8 70.Nd3 Bb6 71.Nc5+ Bxc5 2. ² (0.60): 56.Bd4 Be7 57.Nf4 Kd7 58.Re1 Bc4 59.Be3 Rh7 60.Rb1 Rhh8 61.Bb6 Kc6 62.Bd4 Kd7 63.Re1 Bb3 64.Bf2 Bc4 65.Be3 3. ² (0.60): 56.Bb4 Bc4 57.Nf4 Kd7 58.Bd6 Bb6 59.Rh2 Rh7 60.Bb4 Bc7 61.Re1 Bb3 62.Rd2 Bc4 63.Bd6 Bd8 64.Bc5 Rhh8 65.Ra1 Bc7 66.Bd4 Bd8 67.Kg2 Be7 68.Rc1 Rh7 69.Bb6 Kc6 4. ² (0.60): 56.Ke3 Bc4 57.Bd4 Be7 58.Kf3 Kd7 59.Re1 Rh7 60.Kf4 Kc6 61.Ra1 Rhh8 62.Kf3 Rh7 63.Nf4 Kd7 64.Re1 Rhh8 5. ² (0.60): 56.Re1 Bc4 57.Bd4 Bc7 58.Nf4 Kd7 59.Bc5 Rh7 60.Bg1 Rhh8 61.Bf2 Bd8 62.Ra1 Be7 6. ² (0.60): 56.Rf1 Bc4 57.Bd4 Be7 58.Re1 Kd7 59.Kf4 Kc6 60.Ke3 Rh7 61.Kf3 Bb3 62.Ra1 Kd7 63.Bc5 Bd8 64.Bd6 Bb6 65.Re1 Bc4 66.Nf4 Bd8 67.Rb1 Bb6 7. ² (0.60): 56.Ra1 Bc4 57.Bf2 Be7 58.Nf4 Kd7 59.Be3 Bb3 60.Bg1 Bc4 61.Bd4 Rh7 62.Bb6 Bb3 63.Ba7 Rhh8 64.Ke2 Bc4+ 65.Ke3 Rh7 66.Kf3 Rhh8 67.Re1 Rh7 68.Bd4 Rhh8 69.Bb6 Rh7 70.Bd4 8. ² (0.60): 56.Ba7 Bc4 57.Nf4 Rh7 58.Be3 Kd7 59.Nd3 Bc7 60.Bd4 Kc6 61.Rh1 Bd8 62.Nc5 Re8 63.Rc1 Bc7 64.Nd3 Rg8 65.Rcd1 Rhh8 66.Rh1 Bd8 67.Nc5 Re8 68.Kf4 Rh7 69.Nd3 Be7 70.h5 Bxd3 71.Rxd3 Rxh5 9. ² (0.60): 56.Bg1 Be7 57.Nf4 Kd7 58.Bf2 Bc4 59.Bd4 Rh7 60.Nd3 Rhh8 61.Be3 Rf8 62.Nf4 Rfg8 63.Bb6 Kc6 64.Bd4 Kd7 65.Re1 Rh7 66.Bb6 Rhh8 67.Ba5 Rh7 68.Rg1 Bc5 69.Rh1 Be7 70.Re1 Bc5 71.Rh1 10. ² (0.60): 56.Bf2 Be7 57.Nf4 Kd7 58.Re1 Bc4 59.Bb6 Rh7 60.Kg2 Rhh8 61.Kf2 Bb3 62.Kf3 Bc4 63.Be3 Bd8 64.Bc5 Bc7 65.Nd3 Rh7 66.Bf2 Kc6 67.Rh1 Kd7 68.Bd4 Rgh8 69.Re1 Ba5 70.Nf4 |
|
Jun-13-11
| | kutztown46: Part 9:
Ten moves at 0.60, and Bd6 is not one of them! After forcing 56. Bd6, Stockfish gives the following: After 55...Kc6 56. Bd6:
click for larger viewAnalysis by Stockfish 2.0.1 JA 64bit (33-ply):
1. ² (0.56): 56...Bc4 57.Nf4 Kd7 58.Rbd1 Bb3 59.Rg1 Bc4 60.Ra1 Rh7 61.Nd3 Bb6 62.Nc5+ Bxc5 63.Bxc5 Kc6 64.Bd6 Rgh8 65.Kf4 Bb3 2. ± (1.17): 56...Re8 57.Nf4 Rhg8 58.g4 fxg4+ 59.Kxg4 Bc4 60.Rh1 Rg7 61.Rdh2 Bb3 62.Rf1 Bc4 63.Rff2 Bb3 64.Ne2 Bd1 65.Rf1 Bxe2+ 66.Rxe2 Rh8 67.Bf8 Rgh7 68.Bh6 Kd7 69.Ref2 Be7 70.Kg3 Kc6 71.Rf3 Rg8 72.Kg4 3. ± (1.37): 56...Rg7 57.Nc5 Re8 58.Nxb3 axb3 59.Rd4 Ra7 60.Rb4 Ra4 61.Rxb3 Ba5 62.Bb4 Bc7 63.Kf4 Rg8 64.Rh1 Rh8 65.Rd1 Rh7 66.Ra1 Rh8 67.Rf1 Raa8 68.Rd1 Ra4 69.Re1 4.  (1.41): 56...Rh7 57.Nc5 Re8 58.Nxb3 axb3 59.Rd4 Ra7 60.Rb4 Ra4 61.Rxb3 Ba5 62.Bb4 Bc7 63.Kf4 Rg8 64.Rh1 Rh8 65.Rd1 Kd7 66.Re1 Re8 67.Re2 Rc8 68.Kf3 Rh8 69.Kf4 Rc8 No big deal!
During the game, <AK> kept hoping that Rybka would find a way to advance its h-pawn. I will leave that very interesting part of the analysis to <AK> to see if he can make it work. After 91. b4, which Rybka played apparently to avoid the 50-move rule, I went over the remaining moves carefully, looking for improvements for black. Initially, I thought 98...Rhc8 and 100...Be3 were improvements but they both petered out to 0.00. I'll not show the analysis. |
|
Jun-13-11
| | kutztown46: Part 10:
Here is the Stockfish analysis of the position after 106...Kd7 (at which point the draw was agreed): After 106...Kd7:
click for larger viewAnalysis by Stockfish 2.0.1 JA 64bit (32-ply):
1. ² (0.56): 107.Bb4 Rc2+ 108.Kxg3 Bc7 109.Rxh8 Bxe5+ 110.Kf3 Bxh8 111.a5 Kc6 112.Ra1 d3 113.Ra3 Bd4 114.Rxd3 e5 115.a6 Rc4 116.Rb3 Kd5 117.h5 gxh5 118.Bd2 h4 119.Kg4 Ra4 120.Rb5+ Ke6 121.Ra5 Rc4 122.Kh3 Rc2 2. = (0.00): 107.R1b7+ Kc6 108.Rf7 Rxh4 109.Rxd8 Rh2+ 110.Kg1 Rc1+ 111.Rf1 Rcc2 112.Rc8+ Kd5 113.Rxc2 Rxc2 114.Rd1 Ke4 115.a5 d3 116.a6 Ra2 117.Bb4 Ke3 118.Rf1 Rxa6 119.Kg2 Ra2+ 120.Kxg3 d2 121.Bxd2+ Rxd2 122.Rf6 Rd1 3. = (0.00): 107.Rf1 Rxh4 108.Rf7+ Kc6 109.Rxd8 Rh2+ 110.Kg1 Rc1+ 111.Rf1 Rcc2 112.Rc8+ Kd5 113.Rxc2 Rxc2 114.Rd1 Ke4 115.a5 d3 116.a6 Ra2 117.Bb4 Ke3 118.Rf1 Rxa6 119.Kg2 Ra2+ 120.Kxg3 d2 121.Bxd2+ Rxd2 122.Rf6 Rd1 4. = (0.00): 107.Rh1 d3 108.Bb4 Rc4 109.Rd1 Rhxh4 110.Rxd3+ Rhd4 111.Rxd4+ Rxd4 112.a5 Bc7 113.Rb7 Kc6 114.a6 Rd5 115.Bd6 Bxd6 116.exd6 Kxd6 117.a7 Ra5 118.Rg7 Kd5 119.Kxg3 e5 120.Rd7+ Kc6 121.Re7 Kd5 122.Kf3 Ra3+ I slid forward from that position to see if either side had chances. The best I can tell, the answer is no. Here is the line I found: <107.Bb4 Rc2+ 108.Kxg3 Bc7 109.Rxh8 Bxe5+ 110.Kf3 Bxh8 111.a5 Kc6 112.Ra1 d3 113.Ra3 Rh2 114.Rxd3 Rh3+ 115.Ke2 Rh2+ 116.Kd1 Rxh4 117.Rd6+ Kc7 118.Rb6 Rc4 119.Ke2 Be5 120.Rxe6 Rxb4 121.Rxe5 Kd6> Here is Stockfish's final analysis - yes, that is 50-ply! Final Position Sliding Forward:
click for larger viewAnalysis by Stockfish 2.0.1 JA 64bit (50-ply):
1. = (0.00): 122.Re8 Ra4 123.Rg8 Rxa5 124.Rxg6+ Ke5 125.Kd3 Kf4 126.Kc4 Rxg5 127.Rxg5 Kxg5 2. = (0.00): 122.Re3 Ra4 123.Rd3+ Ke6 124.Re3+ Kd6 |
|
Jun-14-11
 | | Golden Executive: Very deep analysis <kutztown46>, You asked for the chat record of the continuation of the game: <PART 1>
"*** aylerkupp has joined the table.
*** dragon_prz64 has joined the table.
*** kutztown46x has joined the table.
*** goldenexecutive10 has joined the table.
kutztown46x: that should do it.
aylerkupp: Well, that wa relatively painless and quick. kutztown46x: I will comapre positions with my GUI.
aylerkupp: When the Yahoo clock shows 27 mins I'll start Rybka
kutztown46x: positions agree.
aylerkupp: Yeeeah !!!
kutztown46x: does your Arena show the same position?
dragon_prz64: Holy Mackarel, Batman, 69 moves in least than 6 minutes aylerkupp: NO, I have a rook on e8. I will change to e8. kutztown46x: you have 30 minutes.
goldenexecutive10: AK average move: 4.35 seconds !
kutztown46x: you should start your engine at 23 minutes on your Yahoo clock, not 27 minutes. kutztown46x: every time Yahoo changes the ad, my display goes blank. I have to minimize and restore the window. kutztown46x: aggravating.
goldenexecutive10: KT average move was less than 7 seconds ! aylerkupp: OK now.
kutztown46x: good.
dragon_prz64: Kutztown, are you using Firefox? If so, use AdBlock as an add-on. aylerkupp: Aaargh! I thought I checked it before the game!!! kutztown46x: I was slow because I printed out the moves and was crossing them off one at a time. kutztown46x: I am using Google Chrome.
kutztown46x: I had to keep switching between my pen and my mouse. aylerkupp: I have the moves in, what else?, an Excel spreadsheet and was going down a cell for each move. kutztown46x: your method was more efficient.
aylerkupp: You handled that well. Can you juggle 3 things? kutztown46x: if I try, I may repeat GE's flying papers scenario. aylerkupp: Maybe, but it didn't help me get the *.pgn file and the Excel spreadsheet to agree. aylerkupp: Sometimes low tech is best.
kutztown46x: I did not get a chance to post the chat from Saturday yet. aylerkupp: The clock sure moves slowly when you're waiting for it to reach a certain time. kutztown46x: 22 more minutes.
aylerkupp: I think I'll start a book.
kutztown46x: then yu make one move, and I must wait 14 minutes! goldenexecutive10: I am impressed the speed you showed ! kutztown46x: reading a book or writing a book?
aylerkupp: You can start a book also, if you want.
<kutztown46x: dragon, are you going to submit the games to cg?dragon_prz64: Yes, I still have some work to do with the annotations for the last 3 rounds left. goldenexecutive10: I really will miss the action now the tournament is almos finished goldenexecutive10: i mean almost
kutztown46x: a lot of analysis has been posted - can it be transferred to the game pages once they exist? aylerkupp: I can help to reformat the analysis as needed for easier posting if you want. dragon_prz64: Most of the engine analysis, I used page # of AylerKupp forum but most of the notes by the participants can be transferred to the game pages.> |
|
Jun-14-11
| | Golden Executive: <PART 2>
"kutztown46x: AK, I can see why Rybka did not try to push the h-pawn. kutztown46x: once black rook is on h5, what can white do? aylerkupp: Rooks on h2 and e1, then g4. After that, who knows? aylerkupp: There's also a line involving b3, ...axb3, a4 and the q-side is open. kutztown46x: well, maybe. but if g4, then 50-move rule counter starts over! aylerkupp: OK, so it does. More chances for R to win. aylerkupp: I was playing the black pieces and thoughtlessly moved both rooks off the back rank. R played b3
.
kutztown46x: if we do reach 50 moves, I wonder if Yahoo will recognize it.
kutztown46x: For that matter, will my Fritz GUI recognize it? aylerkupp: Yahoo should since we're playing the game from the start. Less sure about our GUIs. aylerkupp: I feel like the soldiers in WW I waiting for the signal to jump out of their trenches. goldenexecutive10: In my last game vs the masters, qilzqhynw posted: "this is a must-win game for the allies right? " What did he mean? aylerkupp: Or the drivers at Le Mans waiting for 4:00 PM for the race to start. aylerkupp: qilzqhynw meant that the Masters (allies) had to win in order to win the tournament. kutztown46x: GE, he meant the humans had to win, or else the best they can do is 2nd place. kutztown46x: only R or S can win the tournament now.
aylerkupp: 5 mins to go for me
kutztown46x: no, you must wait for 23 minutes.
aylerkupp: Oh, I thought it was 27. Then I better reset my GUI's clock.
kutztown46x: I had 37, you had 23.
kutztown46x: You can verify that on your forum.
aylerkupp: Yes, of course. You're right. Another senior moment. goldenexecutive10: Thanks for clarifying !
aylerkupp: I hope I remember to enter the move on Yahoo once Rybka makes its first move. kutztown46x: I'll remind you once your clock goes down to 1 minute. aylerkupp: I cranked up the volume on my speakers just to make sure I heard the move announcement. kutztown46x: If there is a check, it may split your eardrums. aylerkupp: I feel like I'm watching Iron Chef America, Top Chef Masters, or some other cooking show when they count down the time remaining. aylerkupp: Nope, I can't see any legal checks.
kutztown46x: I could play Bxe5+.
aylerkupp: Or a missile launch, 10...9...8...7...
aylerkupp: After the waiting the game itself will probably be anticlimactic kutztown46x: really?
goldenexecutive10: Its like and adjourment game
aylerkupp: I hope I can find the spacebar when the time comes. kutztown46x: 1 minute.
aylerkupp: I'm sure glad I have my diaper on.
aylerkupp: OK, R is calculating.
aylerkupp: And I almost missed it.
kutztown46x: OK, my turn to wait.
goldenexecutive10: I really like the sportsmanship behavior at this tournament aylerkupp: I'll wake you with 1 min to go.
aylerkupp: It's not like it's a life and death matter. kutztown46x: when Yahoo clock gets to about 23:05, I will start my engine and enter your move. aylerkupp: And, besides, it's more fun this way.
kutztown46x: I meant 37:05.
aylerkupp: Yes, I was about to tell you that.
aylerkupp: Talk about good sportsmanship!
kutztown46x: It will take a little time to rebuild the hash table. kutztown46x: on Saturday S wa@@@@@@@@@ 32-ply in seconds. kutztown46x: why did it display that way?
aylerkupp: R was in the low to mid 20's. When you have a large hash table and are repeating moves, that's what happens. kutztown46x: I know why! |
|
Jun-14-11
| | Golden Executive: <PART 3>
""kutztown46x: 6 more minutes.
aylerkupp: I just changed my diaper, just to make sure. aylerkupp: Sorry about my ultra conservative time mis-management. kutztown46x: no problem.
kutztown46x: 1 minute.
aylerkupp: It's getting near your bedtime. How much longer do you want to practice moving the pieces back and forth? *** dragon_prz64 has joined the table.
*** dragon_prz64 has left the table.
kutztown46x: it's like a lullaby.
kutztown46x: it's like a lullaby.
aylerkupp: With a bad melody
*** dragon_prz64 has joined the table.
*** dragon_prz64 has left the table.
aylerkupp: R is actually expecting 85...Bxd4. Dreamer! aylerkupp: This position wasn't in its hash table.
aylerkupp: I sure hope one of the engines doesn't decide to make a capture. *** kutztown46x has left the table.
*** kutztown46x has joined the table.
aylerkupp: Phew! I thought you were REALLY bored.
kutztown46x: disconnected me
aylerkupp: Fast recovery
aylerkupp: Oh no!!!
kutztown46x: what?
aylerkupp: R's evals have been dropping the last few moves. Who knows why? aylerkupp: I think R just committed suicide. Or maybe it was also bored aylerkupp: Butnow it doesn't have much time on its clock. dragon_prz64: AK, move faster
*** Draw requested.
aylerkupp: It's finally interesting.
*** Game ended in a draw.
*** aylerkupp's old rating: 1228; new rating: 1227
*** kutztown46x's old rating: 1219; new rating: 1219
aylerkupp: Thanks.
aylerkupp: Talk about sportsmanship.
aylerkupp: Why not? It's part of the game.
kutztown46x: But I did want to get you in time trouble to see if R would blunder at low ply. aylerkupp: I just wi@@@@@ had decided to make its move much earlier. ---
kutztown46x: final eval was 0.00
aylerkupp: And something very odd happened. It's evals were at +1.02 for many moves and then they began to drop just before it moved its pawn. kutztown46x: let's talk about it later. I need to save the game, unwind, and get to bed. goldenexecutive10: Congratulations for both of you
kutztown46x: save the chat. I lost it all when it disconnected me. *** dragon_prz64 has left the table.
aylerkupp: Good night, sweet dreams, and don't let the bed bugs bite. Lots to talk about. Strange game. *** kutztown46x has left the table.
goldenexecutive10: Thanks thanks thanks
goldenexecutive10: great people here" |
|
Jun-20-11
 | | AylerKupp: Masters vs. Machines, Rybka vs. Stockfish, Jun-04-11 Pre-Game preparations – CCRL 40/40. I decided to take a break from the Team White vs. Team Black game and return to the Masters vs. Machines tournament. I was encouraged by the success of my pre-game preparation for the previous Houdini vs. Rybka game even though Rybka couldn't quite come up with the win. So I did something similar prior the Rybka vs. Stockfish game. I'm including all this trivia in case you find it amusing and shows what can happen when you have a MDOP (Meticulous, Detail-Oriented Person) Excel junkie with obviously too much time on their hands. If you don't find it amusing and/or interesting, well you were warned not to read further. First I downloaded all the latest CCRL 40/40 Rybka vs. Stockfish games. There were a lot more games (246 total, compared with the 41 games between Houdini vs. Rybka). It was also more involved since there were multiple Rybka versions (4, 4.1) and multiple Stockfish versions represented (1.7.1, 1.8, 1.9, and 2.0.1) as well as 64-bit and 32-bit versions of both engines. Since Rybka had White and the choice of opening move, I also looked at their relative record for games starting with 1.e4 and 1.d4 (I consider any other opening moves to be for wimps), as well as Rybka 4.x's record against Stockfish 2.0.1 and the number of moves per game. The results: All Rybka 4.x against all Stockfish versions:
Rybka 4.x wins 91/246 games, 37.0%
Stockfish (all versions) wins 24/246 games, 9.8%
Draws: 131/246 games, 53.3%
Average moves/game = 73
Houdini vs. Rybka showed White with a remarkable 59.2% winning percentage but a 37% vs. 10% winning advantage by Rybka over Stockfish was still very encouraging. All Rybka 4.x against Stockfish 2.0.1 version:
Rybka 4.x wins 25/65 games, 38.5%
Stockfish 2.0.1 wins 5/65 games, 7.7%
Draws: 35/65 games, 53.8%
Average moves/game = 76
So Rybka 4.x actually did slightly better against Stockfish 2.0.1 than against the combined set of all Stockfish versions, and the average game a little bit longer. So, to simplify things, I looked at Rybka's play against all Stockfish versions since this would also give me a larger statistical sample. Looking at 1.e4 vs. 1.d4 I got the following results for Rybka 4.x against all Stockfish versions: After 1.e4 (105/246 = 42.7% of the games):
Rybka wins 48/105 games, 45.7%
Stockfish wins 10/105 games, 9.5%
Draws: 47/105 games, 44.8%
Average moves/game = 74
After 1.d4 (99/246 = 40.2% of the games):
Rybka wins 27/99 games, 27.3%
Stockfish wins 10/105 games, 12.1%
Draws: 47/105 games, 60.6%
Average moves/game = 71
So clearly Rybka does much better against Stockfish with 1.e4 than 1.d4. Which pleases me since I'm a confirmed 1.e4 player. Or, as Bobby Fischer would have said, "Best by test". The average length of a game was also a little bit longer after opening with 1.e4 than 1.d4. |
|
Jun-20-11
| | AylerKupp: Masters vs. Machines, Rybka vs. Stockfish, Jun-04-11 Pre-Game preparations – CCRL 40/40 (continued). Now on to determining what would be the best opening and variation to use. To simplify the comparisons I calculated a Rybka/Stockfish (R/S) performance ratio where each win by an engine was worth 1 point, each draw was worth 0.5 points, and each loss was worth 0 points; then divided one engine's performance by the other engine's performance. Basically, a R/S performance ratio > 1 means that Rybka outperformed Stockfish in a given set of games and a R/S performance ratio < 1 means that Stockfish outperformed Rybka in that set of games. So for 1.e4 openings, the R/S performance ratio was: (48 + 0.5*47) / (10 + 0.5*47) = 2.53 For 1.d4 openings the R/S performance ratio was: (27 + 0.5*47) / (10 + 0.5*47) = 1.36 As expected, the R/S performance ratio for games starting with 1.e4 was much higher than for games started with 1.d4. The following openings gave the indicated R/S performance ratios for various openings when more than 3 games were played with that opening between the engines, with R/S performance ratios in parenthesis: Rybka 4.x vs. Stockfish (all versions):
Opening Games R/S
Philidor 4 (4.00)
Gruenfeld 4 (3.00)
Caro-Kann 9 (2.60)
QGD Slav 5 (2.33)
Ruy Lopez 16 (2.20)
Benko's opening 11 (2.14)
Sicilian 44 (2.03)
QGD semi-Slav 15 (2.00)
Scotch game 3 (2.00)
English 16 (1.91)
Reti 14 (1.80)
French 16 (1.67)
King's Indian 17 (1.43)
Queen's Gambit Declined 12 (1.40)
Queen's pawn game 7 (1.33)
Benoni 3 (1.00)
Neo-Gruenfeld 4 (1.00)
Queen's Indian 13 (0.86)
Nimzo-Indian 11 (0.83)
The R/S=4.00 is somewhat of a misnomer for Philidor's defense since Rybka won all 4 of its games. So, if you're playing Black against Rybka, and Rybka opens 1.e4, don't play the Philidor! Therefore, the best 1.e4 openings against Stockfish in CCRL 40/40, with R/S performance ratios in parenthesis, were: 1. Caro-Kann (2.60)
2. Ruy Lopez (2.20)
3. Sicilian (2.03)
4. Scotch game (2.00)
5. French (1.67)
So, needless to say, I was very happy that <kutztown46> elected to play the Caro-Kann! And, of the various approaches Rybka tried against the Caro-Kann, the advance variation did best; 2 wins, 0 losses, 2 draws (R/S = 3.00). So this was the reason why I selected this variation in our actual game since I wasn't happy with the results of my old favorite, the Panov-Botvinnik attack (1 draw). This was corroborated by the results of the tournament games. |
|
Jun-20-11
| | AylerKupp: Masters vs. Machines, Rybka vs. Stockfish, Jun-04-11 Pre-Game preparations – Practice Tournament Games But I wanted more direct experience so, like in my preparation for Rybka's game against Houdini, I ran a series of 35 tournament games between Rybka 4.1 (White) against Stockfish 2.0.1 (Black) at a 120 min time control using selected 1.e4 openings, trying to guess what defense <kutztown46> would use. Since Stockfish needed to win in order to win the tournament, I expected a Sicilian or even a Ruy Lopez Marshall , given Houdini's less than stellar performance on the White side of a Ruy Lopez Marshal. I really didn't expect <kutztown46> to select a Caro-Kann. The results of the tournament games were not at all what I expected; Rybka won 7/35 games (20.0%), Stockfish 8/35 games (22.2%) and 57.1% of the games were drawn. Huh? Not at all comparable with the CCRL results, and not exactly encouraging. Maybe it was my choice of openings. Here are the results and the R/S performance ratios: Opening Variation R/S
French Winawer, Poisoned Pawn 3.00
Caro-Kann Panov-Botvinnik attack 1.00
Sicilian Various 0.79
Ruy Lopez Marshall, Anti-Marshall 0.69
I used to play the Panov-Botvinnik attack against the infrequent Caro-Kann with good results since the isolated queen pawn suited my attacking style but the tournament games resulted in 4 draws (R/S = 1.00). I guess defensive play has gotten better. And the French Winawer, Poisoned Pawn turned to be a whitewash for Stockfish; 3 loses in 3 games. So, if you're playing White against Stockfish, open 1.e4 and hope for a Winawer Poisoned Pawn. It may be your best chance. On the other hand, the R/S for the Ruy Lopez Marshall, main line was only 0.50 (0 wins, 2 losses, 4 draws) giving credence that engines (at least Houdini and Rybka) don't seem to know how to play the White side of a Ruy Lopez Marshall very well. So I tried various anti-Marshalls (8.a4, 8.d4, and 8.h3) and got a more respectable R/S of 1.00 (1 win, 1 loss, 3 draws). Still not what I had hoped for, but better than the various Marshalls. But I ran out of time to run additional tournament games so I created a short table of those 4 defenses plus other "reasonable" defenses against 1.e4 (Alekhine, Robatch, Pirc, Scandinavian) and various possible lines for each. So I thought I was well prepared and ready to play Stockfish/<kutztown46>. Then the game started . . . |
|
Jul-28-11
| | Golden Executive: <Then, with 1 (one!) move left before a draw via the 50-move rule applied and with 3 min, 1 sec left on its clock, Rybka played 91.b4, restarting the 50-move counter. Disaster! (I had a different expletive in mind, but I censored it)> <What I think happened was that Rybka evaluated it's position as sufficiently better so that a draw was unacceptable, regardless of how much time was left in its clock. The least drastic way that it could avoid a draw was by a pawn move since a piece capture (by a sacrifice) would have driven its evals negative. Since it was a pawn up and the pawn would be regained within a few moves (it may or may not have seen this)> Probably <kutztown46> was in touch with Nakamura, but at the end, <KT> was a gentleman: Rybka vs Nakamura, 2008
Of course this post is just for fun, I have a lot of respect for both of you <AylerKupp> and <kutztown46>. p.s. I also have respect for Nakamura, it´s almost unbelievable he managed that game for a win (blitz game !) |
|
Jul-28-11
| | Golden Executive: I meant this Rybka vs Nakamura, 2008 |
|
Aug-01-11
 | | OhioChessFan: It does my heart good to see a CaroKann played by an engine. Now if I could only convince a Black team to vote for it. 91. b4 Rybka says "I am bored of this Rook shuffling. Let's try something different." |
|
Dec-15-11
 | | Penguincw: I'm surprised there is still this many pieces this late. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 1 of 9. Finally, enough time (probably too much time!) and motivation to comment on the game. That and some well-deserved prodding by <kutztown46>. As usual by now, I set the Arena GUI to 93 minutes/game to give me 12 minutes to deal with Clock Lag. I entered the first 8 moves myself based on my opening cheat sheet rather than allowing Rybka to play the first 8 moves. Again, the game score will be easier to read if you copy it to Word or another word processor, change the spaces to tabs, and adjust the tab setting appropriately so that the moves and stats line up in columns. <kutztown46>'s choice of the Caro-Kann was a surprise to me since I expected him to press for a win since Stockfish needed a win in order to win the tournament. But his philosophy and approach was different as he indicated earlier. I wanted to win to guarantee first place in the tournament since the Masters were able to beat Houdini in their last game with Black and there was a possibility that they could do so again with White. I chose the advance variation because it provided Rybka with the best results in the latest CCRL 40/40 tournament as described and I wasn't satisfied with the winning chances in either the typical main line Caro-Kann or my earlier favorite, the Panov-Botvinnik attack. And my choice of the advance variation was apparently a surprise for <kutztown46> White Black Eval Depth Time
1. e4 c6 - - -
2. d4 d5 - - -
3. e5 Bf5 - - -
4. Nf3 e6 - - -
5. Be2 Ne7 - - -
6. O-O h6 - - -
7. Nbd2 Nd7 - - -
8. Nb3 a6 - - -
click for larger viewTime to let Rybka and Stockfish go at each other. The position after 8.Nb3 was part of a line arrived at by transposition which, after 8...Bh7 provided a 50% winning advantage for White per Opening Explorer so I was satisfied with that.
<kutztown46> chose 8...a6 because it provided the best results for Black but with only 5 games in the database I don't consider that a meaningful statistical sample. And apparently Rybka was also satisfied with the position since it evaluated it slightly in White's favor. I let Rybka take its time before making its next few moves to make sure that it didn't mess up the good position that I had given it ļ but time management is Rybka's Achilles' Heel; it takes a long time in the opening when it has time on its clock, typically leaving itself with less time on its clock at the later part of the game as compared with either Houdini or Stockfish. <kutztown46> knew this and planned to try to take advantage of it. But I figured that it was still early in the game and that I could make up the time differential later when Rybka's hash table was full and it could get to deeper depths quicker. But, as we all know by now, this didn't exactly work out well. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 2 of 9. White Black Eval Depth Time
9. Bd2 Ng6 +0.31 19 201
10. Ne1 h5 +0.40 18 235
11. h3 c5 +0.41 18 143
click for larger viewThis I found to be the most interesting move by Stockfish. It gives up a pawn but in return it gets the two bishops plus an attack. I mentioned that it reminded me of a Ruy Lopez, Marshall Attack when Black does a similar thing although the follow-up is, of course, much different. <kutztown46> indicated that Stockfish preferred 12.Be3 over 12.Bxh5. If given more time (almost 9 mins in Infinite Analysis mode) Rybka at d=17 prefers 12.Bxh5 over 13.Be3 by a small margin, [=0.66] vs. [+0.54]. But given an additional 5 mins and one more ply Rybka now prefers both 12.c3 {+0.54] and 12.dxc5 [+0.50] over both 12.Bxh5 and 12.Be3, both evaluated at [+0.48]. The game would have taken a very different turn if Rybka had chosen either of those 2 alternatives. White Black Eval Depth Time
12. Bxh5 Qh4 +0.39 16 170
13. Bxg6 Bxg6 +0.46 17 109
14. Nf3 Qh5 +0.38 19 172
15. Nxc5 Nxc5 +0.38 17 35
16. dxc5 Bxc5 +0.38 18 0
17. c3 - +0.38 17 5
click for larger viewIn return for his pawn Stockfish has good squares for his bishops and Rybka has locked in his bishop slightly with 17.c3. Maybe 17.Bc3 preserving the bishops freedom of movement and overprotecting the Pe5 was better, but then White has to worry about defending its Pc2. Rybka calculated for only 5 secs before selecting 17.c3 so it must have decided on it while it was spending all that time on moves 9 – 14. After 17.Bc3 Black also has the possibility of 17...Be4 followed by 18...Bxf3 19.Qxf3 Qxf3 20.gxf3 Rxh3 regaining the pawn with the better position, so maybe this was what Rybka was trying to prevent. Instead, Stockfish's next move forces Rybka's KR to go where it probably wanted to go anyway. White Black Eval Depth Time
17. ... Bd3 - - -
18. Re1 Qg6 +0.39 18 0
19. Ng5 - +0.40 18 17
click for larger view<kutztown46> indicated that Stockfish preferred 19.Qa4+ followed by 20.Qg4 and exchanging queens rather than Rybka's 19.Ng5. As you can see below Rybka took only 17 secs to select 19.Ng5 so it must have also considered it during its earlier calculations. If given more time (a little over 10 mins at d=20 in Infinite Analysis mode) Rybka still would have selected 19.Ng5 over 19.Qa4+ by a narrow margin, and this how it thought the game might continue: [+0.44]: 19.Ng5 Be7 20.f4 Kf8 21.Be3 Kg8 22.Qd2 Rc8 23.Red1 Bf5 24.Bd4 Bc2 25.Re1 Rh5 26.Be3 Rh6 27.Rac1 Bd3 [+0.42]: 19.Qa4+ b5 20.Qg4 Qxg4 21.hxg4 Kd7 22.Be3 Be7 23.Nd4 Bg6 24.f3 Bh4 25.Bf2 Be7 26.Rad1 Rab8 27.Re2 Rh7 28.Bg3 Rhh8 29.a3 Kc7 |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 3 of 9. White Black Eval Depth Time
19. ... Rc8 - - -
20. Qf3 Be7 +0.42 18 213
21. Qg3 Rh5 +0.42 17 0
22. Nf3 Qxg3 +0.47 19 3
click for larger viewRybka considers the position after the exchange of queens as advantageous for White, and its eval started to climb, even though White extra pawn is now doubled, its Pe5 isolated, and it's hard for White to make use of the half-open f-file because of Black's LSB's command of the a6-f1 diagonal. Which Black immediately surrenders! White Black Eval Depth Time
23. fxg3 Be4 +0.52 15 14
24. g4 Rh8 +0.58 18 70
25. Ng5 Bd3 +0.58 17 85
click for larger viewBack again on the a6-f1 diagonal. Rybka now decides to either improve it's king's position or expose its king, depending on how you look at it. White Black Eval Depth Time
26. Kh2 Rc4 +0.63 19 156
27. Kg3 Kd7 +0.63 17 36
28. Bf4 b5 +0.64 17 8
click for larger viewSetting up the usual minority attack in these types of positions, but Stockfish doesn't follow through. In spite of Black's pawn minus I prefer Black's position here. The minority attack will open up lines on the q-side and Stockfish should be able to make good use of its two bishops. White has no play on the k-side due to its doubled pawns and inability to use the half-open f-file. Still, Rybka continues to prefer its position and its eval continues to climb. Note that 28.Nxf7 loses the exchange to 28...Bh4+, something I didn't notice at the time. White Black Eval Depth Time
29. a3 Bc2 +0.62 19 152
30. Nf3 Be4 +0.62 18 104
31. Be3 Bc2 +0.69 17 66
32. Bd4 Rcc8 +0.72 17 34
33. h4 f5 +0.79 18 208
34. g5(!) - +0.80 18 71
click for larger viewA good move by Rybka, I think, keeping the position closed. Either 34.gxf5 Bxf5 or 34.(any) fxg4 would have allowed Stockfish to open up the position somewhat and all of a sudden it's Black that has play on the f-file against White's suddenly exposed king. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 4 of 9. White Black Eval Depth Time
34. ... Be4 +0.80 18 71
35. Be3 g6 +0.80 19 12
36. Rh1 Bd3 +0.81 19 83
37. Kf4 Be4 +0.89 18 127
38. g3 a5 +0.90 18 0
39. Rhf1 Bd3 +0.90 19 95
click for larger viewHere I thought that perhaps 39...Bxf3 might have been better, Black getting rid of its bad bishop in exchange for White's knight which ought to be better in a closed position. Black would then be left with a good DSB vs. White's bad DSB. But Rybka assessed the position after 39...Bxf3 and either 40.Rxf3 or 40.Kxf3 at [+0.95], d=21, effectively the same as the game continuation. And Stockfish 2.2.2 in infinite analysis mode at d=26 didn't consider either 39...Bxf3 or the game's 39.Bd3 among its top 5 PVs, and evaluated the position at [+0.44] for all 5 PVs. White Black Eval Depth Time
40. Rfd1 Be2 +0.93 19 15
41. Rd2 Bc4 +0.93 17 31
42. Nd4 a4 +0.93 20 12
click for larger viewI thought that Stockfish's 42...a4 was an error, giving up the possibility of a minority attack, since now the position is entirely closed and Black has no prospects of q-side play without a pawn sac or two. White, on the other hand, theoretically has either g3-g4 or h4-h5 pawn breaks after suitable preparation so now the initiative should be in White's hands. Assuming, of course, that Rybka could figure this out. And it couldn't, so instead it goes into a series of meaningless moves which do nothing to advance its position. I call this phenomenon "engine dithering". And this was the last pawn move for quite a while although at the time neither <kutztown46> nor myself appreciated the significance that the game would be drawn following Black's 92nd move due to the 50-move rule. White Black Eval Depth Time
43. Ne2 Rb8 +0.93 21 0
44. Kf3 Rbg8 +0.93 22 72
45. Bd4 Bd8 +0.93 23 0
46. Nf4 - +0.93 23 52
click for larger viewHere Rybka missed a chance to activate its bishop by 46.Bc5, although it succeeds in doing so later. White's DSB would be more active in the a3-f8 diagonal, and any exchange of DSBs should favor White since it would have a N vs. bad B in a closed position. Unfortunately the attempt to force h4-h5 after Stockfish's 46...Be7 by either 47.Rh2 or 47.Rh1 fails against 47...Bxg5! since either rook move pins the Ph4. But even after a laborious sequence such as 47.Rdd1 Be7 48.Rh1 Rh7 49.Rh3 Bd8 50.Rah1 Bb3 51.Kg2 Bc2 52.R1h2 Be4+ 53.Kf2 Bd7 54.h5 leads to no more than equality (Stockfish, [0.00], d=30) or a miniscule edge for White (Rybka, [+0.19], d= 20) after 54...Bxg5 55.hxg6 Rxh3 56.Rxh3 Bxf4 57.gxf4 Rxg6. And, with a completely closed position and opposite colored bad bishops for each side, Stockfish's assessment seems more accurate. So there doesn't look like trying to force h4-h5 gives White any winning chances. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 5 of 9. White Black Eval Depth Time
46. ... Be7 +0.93 23 52
47. Rad1 Bb3 +0.93 23 0
48. Rf1 Rh7 +0.93 24 57
49. Rc1 Bc4 +0.93 24 77
50. Rcd1 Rhh8 +0.93 23 4
51. Kg2 Bb3 +0.93 24 74
52. Rb1 Bd8 +0.93 25 89
53. Nd3 Be7 +0.94 22 62
54. Bc5 - +0.94 22 15
click for larger viewFinally Rybka figures out that it would be advantageous to exchange DSBs. Of course, Stockfish figures out the same thing and refuses the offer to exchange, but now White has its DSB in a more active diagonal with the d4 square free for it's knight. But Rybka doesn't consider either of these as an advantage so its eval of the position is unchanged. And, as you'll see below, it can't open up any files on the k-side as long as Stockfish has its DSB pointed at the Pg5 so it doesn't have anything meaningful to do with its rooks. White Black Eval Depth Time
54. ... Bd8 +0.94 22 15
55. Kf3 Kc6 +0.94 21 3
56. Bb4 Bc4 +0.93 22 137
57. Ne1 Bb6 +0.97 23 66
58. Nc2 Kd7 +0.97 24 41
59. Nd4 Bc7 +0.97 25 65
60. Kf4 Re8 +0.97 21 7
61. Rc1 Rh7 +0.97 24 58
62. Rcd1 Rhh8 +0.97 24 79
63. Bc5 Rc8 +0.97 24 100
64. Rc1 Rce8 +0.97 22 0
65. Re1 Rh7 +0.97 23 77
66. Rh1 Reh8 +0.97 20 6
67. Rhd1 - +0.97 23 83
click for larger viewDuring the game I thought that Rybka could perhaps try either g3-g4 or h4-h5 to open up the position on the k-side. After 67.Rdh2 Rh5 68.Nf3 Black can no longer prevent g3-g4. But after 68...Re8 69.g4 fxg4 70.Kxg4 Bd3 Stockfish evaluates the position at [0.00], d=24 and Rybka evaluates the position at only [+0.29], d=18, quite a comedown from its current [+0.97] eval so it doesn't look like my g3-g4 idea had any merit. White Black Eval Depth Time
67. ... Rd8 +0.97 23 83
68. Rg2 Re8 +0.97 22 0
69. Rh2 Rh5 +0.97 21 34
click for larger viewHere the game was "adjourned" after Rybka made the illegal move 70.h5 since it wasn't told (by me) that Black had played 69...Rh5. But at least it vindicates my judgment that the h4-h5 pawn break was desirable even if unachievable! After the game was resumed 2 days later, Rybka's eval of the position increased from [+0.97] to [+1.02], another indication that multi-core chess engines are non-deterministic. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 6 of 9. White Black Eval Depth Time
70. Rhd2 Rh7 +1.02 23 56
71. Re1 Reh8 +1.02 23 52
72. Bd6 Bb6 +1.02 22 40
click for larger viewAnother attempt by Rybka to induce Stockfish to exchange DSBs which Stockfish once again declines since 72...Bxd6 73.exd6 Kxd6 fails to 74.Rxe6+ and 75.Rxg6. So Rybka once again resorts to "meaningless rook moves" in its "engine dithering" mode. And each time that Rybka tries to force h4-h5 by doubling rooks on the h-file Stockfish prevents it by doing the same thing. But, with a pawn advantage, Rybka doesn't want to repeat the position 3 times. White Black Eval Depth Time
73. Rh1 Ba5 +1.02 25 27
74. Bb4 Bb6 +1.02 22 41
75. Re1 Rc8 +1.02 22 42
76. Rh1 Rch8 +1.02 24 32
77. Rc1 Re8 +1.02 23 36
78. Rg1 Bc7 +1.02 24 41
79. Rh1 Reh8 +1.02 24 52
80. Rhd1 Bb6 +1.02 22 28
81. Rh2 Rh5 +1.02 21 35
82. Rhh1 R5h7 +1.02 22 30
83. Rc1 Rh5 +1.02 24 57
84. Rhd1 R5h7 +1.02 23 31
85. Bd6 Rg7 +1.02 21 19
86. Rd2 Rgh7 +1.02 23 18
click for larger viewNow for some unknown reason Rybka's evals started to fluctuate even though the position's characteristics had not really changed. Little did I know at the time that this may have been an indication of impending disaster. White Black Eval Depth Time
87. Rb1 Ba2 +0.78 17 32
88. Rh1 Bc4 +1.02 18 20
89. Rc1 Rh5 +0.84 17 29
90. Rb1 Rc8 +0.65 18 25
91. b4 - +0.16 18 20
click for larger viewAnd now, with a pawn advantage, with about 4 ½ minutes on its clock after its 90th move and only 2 more moves left before the game is declared a draw by the 50-move rule, I think that Rybka determined that a draw was unacceptable given its favorable evaluation of its position. So, in what may have been an unusual example of the horizon effect, Rybka decides to both move and sacrifice a pawn to prevent the 50-move rule draw with the least negative impact on its evaluation of the position (it's only other available pawn move, 91.g4, would have resulted in a significant advantage for Black after 91...Rxh4, and any sacrifice of the knight would presumably have been even worse). Note that after the pawn sacrifice Rybka's evaluation of the position is still favorable although, of course, not as favorable as it had been when it had a pawn advantage. I can only guess that if Rybka's eval of the position had been negative (unfavorable) then it would have preferred the draw by the 50-move rule. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 7 of 9. White Black Eval Depth Time
91. ... axb3 +0.16 18 20
click for larger viewAt this point Rybka had about 3 minutes left in its clock. Normally it would not have been a problem for Rybka to play another 50 moves in the time remaining (making the game essentially a blitz game) although, of course, it could not have calculated as deeply and the quality of its play would have suffered. But, because I had set Ponder=ON, Rybka would have been able to use Stockfish's calculation time for its own analysis and so would likely have been able to play at a reasonable level (i.e. better than me). But the game circumstances were such that there was both a dummy in the loop (me) and a communications lag for the moves to be transmitted between opponents, adding about 10 secs/move. So Rybka didn't know that it had just lost the game on time. White Black Eval Depth Time
92. Kf3 Ba5 +0.05 20 53
93. Bb4 Bc7 +0.09 22 2
94. Re1 Bb6 +0.09 20 1
95. Rb2 Rhh8 +0.09 18 1
96. Nxb3 - +0.28 13 2
click for larger viewAnd, to make matters even worse given the short amount of time that Rybka has left in its clock, Rybka now decides to open up the position in order to recapture the sacrificed pawn. So all the positions that might have remained in its hash table if the position had remained closed get flushed, and Rybka has to start the analysis anew. So the depth that it is able to calculate drops tremendously, and at low search plies Rybka is no match for Stockfish, even though it still thinks its position is favorable. Stockfish, on the other hand, thinks that its position is favorable after 96...Bxb3, [-0.28], d=26. So we have 2 "happy" engines. Interesting. White Black Eval Depth Time
96. ... Bxb3 +0.28 13 2
97. Rxb3 Rc4 +0.27 15 1
98. Bd6 Kc6 +0.06 18 2
99. Reb1 f4 +0.27 12 3
100. Kg2 Bd8 +0.36 11 6
101. Rxb5 Rxc3 +0.53 12 5
102. R5b3 Rc2+ +0.30 18 2
103. Kf3 fxg3 +0.36 19 1
104. Rb8 - +0.30 20 1
click for larger viewHere Rybka decides to complicate the game even further. Had it played it safe with 104.Kxg3 then the best that Stockfish was able to find for itself (d=30) was 2 draws by repetition, 104...Ba5 105.Rb5 Be1+ 106.Kf3 Rf2+ 107.Kg4 Rg2+ 108.Kf3 Rf2+ and 104...Rc4 105.Rb8 Rcxh4 106.Rxd8 Rh3+ 107.Kg2 Rh2+ 108.Kg1 Rh1+ 109.Kf2 R1h2+ 110.Kg1. Had this been played, I might have been able to make the required moves in the time available, and the game would have ended in a legitimate draw. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 8 of 9. White Black Eval Depth Time
104. ... Rc3+ +0.30 20 1
105. Kg2 d4 +0.30 20 1
106. a4 Kd7 +0.29 19 0
107. Bb4 * +0.57 10 0
click for larger viewHere (107.Bb4 was the actual last move plated by Rybka) <kutztown46> graciously offered a draw for the second time and this time I accepted with about 2 seconds left on Rybka's clock. I still have mixed feelings about it because I feel that 91.b4 is an issue with Rybka's move determination function so, if this move would have caused Rybka to lose on time, then that would have been a legitimate way for Rybka to lose. On the other hand, Rybka had no way of knowing the approximately 10-sec/move operator/communication lag under which the game was being played. Without this lag, in the 3 seconds remaining on its clock it could have blitzed many more moves of reasonable quality, given that it could calculate during the time that Stockfish was calculating. And, as <kutztown46> indicated, after 107.Bb4 Rybka retained some advantage, with Stockfish evaluating the position at [+0.56], d=30 after 107...Rc2+ 108.Kxg3 (given the upcoming 109...Bxe5+ this may not have been best) 108...Bc7 109.Rxh8 Bxe5+ 110.Kf3 Bxh8 111.a5 Kc6 112.Ra1 d3 113.Ra3 (maybe 113.a6 is better; e.g. 113...Bd4 114.a7 Bxa7 115.Rxa7 d2 116.Bxd2 Rxd2 117.Rg7) 113...Bd4 114.Rxd3 e5 115.a6 Rc4 116.Rb3 Kd5 117.h5 gxh5 118.Bd2 h4 119.Kg4 Ra4 120.Rb5+ Ke6 121.Ra5 Rc4 122.Kh3 Rc2. Rybka agreed with Stockfish, with this being its evaluation of the position after 107.Bb4 at d=26: 1. [+0.67]: 107...Rc2+ 108.Kxg3 Bc7 109.Rxh8 Bxe5+ 110.Kf3 Bxh8 111.a5 Kc6 112.Ra1 d3 113.Ra3 Rh2 114.Rxd3 Rh3+ 115.Ke2 Rh2+ 116.Kd1 Rxh4 117.Rd6+ Kc7 118.Rb6 Rc4 119.Bd2 Kd7 120.Rb8 Bc3 121.a6 Bxd2 122.a7 Ra4 2. [+1.32]: 107...Rc8 108.Rxc8 Kxc8 109.Rc1+ Kd7 110.Kxg3 Bc7 111.Rc5 d3 112.a5 Rb8 113.Bc3 Rb3 114.Kf4 Ra3 115.Ke4 Ra4+ 116.Kxd3 Rxh4 117.Kc2 Bd8 118.Bd2 Rd4 119.Kc3 Rh4 120.Kc2 Rd4 121.Kc3 Rh4 122.Kc2 Rd4 3. [+2.83]: 107...Rc4 108.Ba5 Rxa4 109.R1b7+ Kc6 110.Bxd8 Rxh4 111.Kxg3 Rh8 112.Be7 Rxb8 113.Rxb8 Kd5 114.Rb6 d3 115.Kf3 d2 116.Rd6+ Kxe5 117.Rxd2 Ra7 118.Bd6+ Kf5 119.Bc5 Rh7 120.Be3 Rh3+ 121.Ke2 Rh8 122.Rd4 e5 So Stockfish would still have had to be careful, although by sliding forward <kutztown46> indicated that after 107...Rc2+ 108.Kxg3 Bc7 109.Rxh8 Bxe5+ 110.Kf3 Bxh8 111.a5 Kc6 112.Ra1 d3 113.Ra3 Rh2 114.Rxd3 Rh3+ 115.Ke2 Rh2+ 116.Kd1 Rxh4 117.Rd6+ Kc7 118.Rb6 Rc4 119.Ke2 Be5 120.Rxe6 Rxb4 121.Rxe5 Kd6 the position is even after either 122.Re8 Ra4 123.Rg8 Rxa5 124.Rxg6+ Ke5 125.Kd3 Kf4 126.Kc4 Rxg5 127.Rxg5 Kxg5 or 122.Re3 Ra4 123.Rd3+ Ke6 124.Re3+ Kd6. And, indeed, after 122.Re8 Ra4 123.Rg8 Rxa5 the position is a draw per the Nalimov tablebases. |
|
Feb-23-12
| | AylerKupp: Game Score and comments: Rybka 4.1 (AylerKupp) vs. Stockfish 2.0.1 (kutztown46), Jun-04-11 and Jun-06-11, Caro-Kann defense, Advance Variation, part 9 of 9. Thus ended probably the most controversial and drama-filled game of the 2011 Masters vs. Machines tournament, as well a probably one of the least interesting. But what might the actual game conclusion have been under a "real life" (non-analysis) situation? What might have happened if Rybka had tried b2-b4 much earlier than move 91 when it would have had much more time on its clock? To try and find out, I ran a mini-tournament (5 games, 60 minute time control) starting with the position after 107.Bb4 with Rybka 4.1 playing White and Stockfish 2.0.1 (the version that <kutztown46> was using at the time) playing Black. Here are the results. I won't bother showing the moves since they don't look all that interesting and it would make an already insufferably long post even more insufferable long. Game 1: Drawn after 189 moves (82 additional moves) due to 50-move rule. Game 2: Drawn after 198 moves (91 additional moves) due to 50-move rule. Game 3: Drawn after 149 moves (42 additional moves) due to insufficient material. Game 4: Drawn after 183 moves (76 additional moves) due to 50-move rule. Game 5: Drawn after 217 moves (110 additional moves) due to 50-move rule. So it looks like a draw would have been the most likely outcome if you discount the ineptness of Rybka's operator. |
|
Feb-23-12
| | kutztown46: <AK> I just skimmed through your analysis. It's very interesting and very well done. It's been a long time since I looked at this game. <I won't bother showing the moves since they don't look all that interesting and it would make an already insufferably long post even more insufferable long.> LOL. |
|
Mar-07-12
| | Golden Executive: <AylerKupp> I agree with <kutztown46>, your analysis and comments are very good. ---
<69. Rh2 Rh5 +0.97 21 34
click for larger view<Here the game was "adjourned" after Rybka made the illegal move 70.h5 since it wasn't told (by me) that Black had played 69...Rh5. But at least it vindicates my judgment that the h4-h5 pawn break was desirable even if unachievable!>. lol !
<After the game was resumed 2 days later, Rybka's eval of the position increased from [+0.97] to [+1.02], another indication that multi-core chess engines are non-deterministic.> After 62..Rh5:
Aylerkupp / Rybka - Kutztown46 / Stockfish, CG.com Masters vs. Machines Invitational 2011
click for larger viewAnalysis by Houdini 1.5a w32:
70.Bb4 Bd8 71.Rhd2 Bc7 72.Re1 Rhh8 73.Rdd1 Rh5 74.Bd6 Bb6 75.Rh1 Bd8 76.Rhg1 Kc8 77.Nc6 Kb7 78.Nxd8+ Rxd8 79.Rd2 Rh7 80.Rf2 Rdh8 81.Rd1 Kc6 82.Rdd2 Kb6 83.Rd4 Rc8 84.Rd1 Rf7 <85.h5> (Houdini agrees with you).
<  (0.36) Depth: 29/54 > ---
<.... on the other hand, Rybka had no way of knowing the approximately 10-sec/move operator/communication lag under which the game was being played> And <kt46> knew that. I applause his decision offering you a draw. The good sportsmanship was present in the whole tournament. |
|
Dec-23-19
| | AylerKupp: <Stockfish's pawn sacrifice 11...c5> After 11...h3 the game reached the following position:
click for larger view<kutztown46> indicated that Stockfish 2.0.1 evaluated the position after 11.h3 at d=29 and after 4½ mins of calculation at [+0.28] (equal) after 11...c5, [+0.36] after 11.Rc8, and [+0.44] after 11...h4. But it considered White's best reply after 11...c5 to be 12.Be3 so at this point it did not consider 11...c5 to be a pawn sacrifice that should be accepted! For comparison, the current Stockfish 10 at d=29 evaluates the position after 11...c5 at [+0.50] at the threshold of While having a (very) slight advantage. And it still considers 12.Be3 to be White's best reply to 11...c5 so it also does not consider 11...c5 to be a pawn sacrifice that should be accepted. After In contrast, Rybka evaluated the position after 11.h3 at d=17 and approximately 4½ mins of calculation at [+0.34] (equal) after 11...h4, [+0.40] (equal) after 11...c5, and [+0.45] (equal) after 11...Qc7. So it did not consider 11...c5 to be Black's best move. I let Rybka calculate longer (about 2 hours) and at d=22 it still considered 11...h4 to be Black's best reply ([+0.37]) with 11...c5 ([+0.39]) and 11...Rc8 ([+0.40]) to be its next 2 best moves. So additional calculation did not substantially change Rybka's evaluation of the position after 11...h3. It therefore seems that after 11.h3 both Rybka and Stockfish considered 11...c5 to give Black good practical chances for the sacrificed pawn, if accepted. The position clarified somewhat after 11...c5 12.Bxh5 Qh4 13.Bxg6 Bxg6 14.Nf3 Qh5 15.Nxc5 Nxc5 16.dxc5 Bxc5
click for larger viewBlack clearly has compensation for the pawn in its 2 bishops and more active pieces. But is it enough? Stockfish clearly evaluated it to be enough for an advantage and Rybka equally clearly evaluated it as not being enough since it was willing to go into this continuation. And equally clearly neither Rybka nor Stockfish could calculate that Black would necessarily win the pawn back since it still had not done so 74 moves later. Rybka was a pawn up and it evaluated its position as favorable at d=21 and about 10 minutes of calculation after 17.Be3. But wait, didn't Rybka play 17.c3 in the game? Yes it did. But from d=11 to d=21 it considered <either> 17.Be3 or 17.c3 as White's best move; 17.Be3 was considered the best move 6 times and 17.c3 was considered the best move 5 times. I no longer have the data I recorded during and soon after the game so I don't know what depth Rybka had reached when it played 17.c3 but it's clearly somewhere between d=15 (after 34 secs of calculation ) and d=19 (after about 7 mins of calculation). Chess engines evaluate the position after every play and might change its determination of the best move after each of them, so the move they actually play is the move that the engine considers the best move at the time that its time management function says "Move!" In contrast Stockfish 10 (I no longer have access to Stockfish 2.0.1) considered 17.Be3, 17.c3, and 17.Rc1 to be its top 3 moves since d=16 and 1 sec of calculation, and 17.Be3 or 17.c3 to be its top 2 moves since d=16 and 1 sec of calculation. At d=34 and almost 7 mins of calculation it considered 17.c3 to be the White's move with an evaluation of [+0.84] (White stands slightly better) and 17.Be3 to be White's 2nd best move with an evaluation of [+0.65]. From d=16 and d=34 17.c3 was considered White's best move 15 times and 17.Be3 was considered White's best move 4 times. So after some forward sliding both Rybka and Stockfish still considered 11...c5 to give reasonably good practical chances, with White standing slightly better in spite of its pawn advantage. |
|
 |
|
|
< Earlier Kibitzing · PAGE 3 OF 3 ·
Later Kibitzing> |
|
|
|
Kibitzing
