Page 4 of 4

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Fri Feb 04, 2011 9:46 am
by BB+
Massive turn-around in the endgame -- Rybka now evals at -1.25 (as does Houdini), as it looks like Black has the chances with RRB+3 vs RR+6. One question here is whether the engines understand that (for instance) RB vs RPP is "dominated" by RB vs R, in that if the RB-side can't win with the pawns off, there is no win. Maybe even RRB vs RR (+pawns). There's also a few lines that lead to White having 4 (or 5) pawns for a Rook. All quite delicate.

I think White's 45/46th move (Re6, which allowed d2, and then Re2) was poor, and that something like f5 was likely superior to head for a draw.

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Fri Feb 04, 2011 12:41 pm
by BB+
Game 29 sideline: what happens if Black blunders with 26... Re7? [Rybka had Re7 in her PV at move 25].
As usual, most combos only appear in the notes.

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Fri Feb 04, 2011 1:34 pm
by BB+
Game 29, move 31:Rybka played Kg7. I think Qe7 draws more easily (and h5 probably transposes to the Qe7 line).
31... Qe7 32. Bg3 h5 33. Qxd6 Qxd6 34. Bxd6 Ne4 35. Bb8 a6 36. Bc7 Kf8 37. Bxb6 Ke7, and White's extra pawn(s) don't do much (Black can play Nxc3, or Nd2/Nxc4 and try for Kd6/Kxd5, for instance -- the White King is also out of play for now).
31... Qe7 32. Qxh6 Qe5+ 33. Bg3 Qxf5, and Black's kingside pressure (after Ne4) is sufficient to draw. E.g. 34. h4 Ne4 35. Bxd6 Qf2.
Both sides are now at +0.2 after 31... Kg7 32. Bg3 Qc8, whereas Houdini (after a long think before exchanging Rooks at moves 29 and 30) had dropped to -0.01 before Rybka chose Qe7.

The adjudication rules must be improved further!

Posted: Fri Feb 04, 2011 7:18 pm
by MichaelIsGreat
To Martin, the organizer of this great Elite Match Season 1.

Here in this position, the game stops when there was plenty of game left in the game I could say! This match should have lasted much longer and end up in a position that is much clearer than this one. Therefore, I really see a need to improve further your adjudication rules for the next Elite Match. You will have plenty of time to refine them to your liking, as the Elite Match is held only at the end of the Division matches.
[Event "TCEC - Elite Match - S1"]
[Site "http://www.tcec-chess.org"]
[Date "2011.02.04"]
[Round "30.1"]
[White "Rybka 4.0"]
[Black "Houdini 1.5a"]
[Result "1/2-1/2"]
[FEN "7k/p1bb1p2/7p/4p2q/4P3/5NP1/P1Q1B2P/7K b - - 10 44"]
[Variant "normal"]

And here is the full game:


Having said that, great matches at your TCEC Elite Match.


Best Regards
MichaelIsGreat

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Fri Feb 04, 2011 7:23 pm
by Martin Thoresen
Michael,

Draw rules that work 100% of the time is hard to implement.

In this case both engines eval was within 0.05 for 10 consecutive plies,
without any pawn advances or captures and thus the game is drawn.

If any of the two engines had thought there was something
in the position their eval wouldn't be as low, would it?

Again, it's really hard to make the draw rules perfect. If I relax the
settings even more, other games could go on for another 50 moves where
they really should be drawn, and that isn't optimal either...

Best,
Martin

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Sat Feb 05, 2011 7:14 am
by BB+
Game 32 was a nice technical win by Houdini in a rook and bishops ending. Eventually the b-pawn got pushed, and finis.Black has many moves than win on move 68, but I was hoping for (say) 68...Bh8 69. Ke3 Rxd3+ [here again other moves win, such as Bf3+ first] 70. Kxd3 Bf3, and White is helpless. [In some of the game sidelines, there might be a few traps when RBP vs BB doesn't win if White can blockade the pawn on the right-coloured square, though I can't construct a specific such problem that looks that applicable here].

Again it's often hard to provide any trenchant real-time comments, but 61... Bf3 might be better than the final b-pawn push. After 62. Rb1 Rc2 63. Bg6 [here Kd2 is met by forcefully Ba5 it seems] Be3 64. Bxe3 Rxe3 65. Kf2 Rc3 (if nothing else, Black moves the King up, and Ra1-a6 for White does not work to prevent this), Black should be able to win both the White pawns in exchange for the b-pawn -- from my point of view, being 2 pawns up looks to be a cleaner win that being one pawn up and trying to force the b-pawn home, though of course Houdini's method worked too. Note that all the various 2-pawns-up endgames with either Rooks or Bishops should be wins (as the pawns are separated by 2 files).

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Mon Feb 07, 2011 2:52 am
by BB+
Game 39, rather fast decay of Black's position:Score of +0.35 after White's 54th move, Black resigned after move 62. Humans instinctively "just know" that the h4-push here is going to lead to danger for Black, but not so for computers, I guess. note also that there were three consecutive decisive games (Rybka had a pedagogical win with a queenside breakthru and passed b-pawn in Game 38, while she seems not to have calculated deeply enough to avoid 40... Nd5 in Game 37 -- again this was a rather deep tactical sequence, so maybe Black was already in trouble -- EDIT: Norway's 2nd largest newspaper seems to have chosen a position from this sequence in their article about TCEC), when in the first half these openings were part of a long draw string.

Game 40 has Rybka with a mild advantage after an exchange sac to obtain a passed pawn.

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Mon Feb 07, 2011 3:43 am
by BB+
Game 40 has Rybka with a mild advantage after an exchange sac to obtain a passed pawn.
Again it seems that the evaluation differences (as opposed to something like pruning details) between the engines are displayed, as Houdini has blockaded the passed pawn, and is now using the material advantage.
With move 42, Houdini jumped from -0.76 to -2.37 or so. It was a nice process of running the King out of the way (Kd7-c8-b8) to allow the Rook on g8 to finally come into play with great effect.

Re: Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Seas

Posted: Mon Feb 07, 2011 4:25 am
by Martin Thoresen
The match has now concluded and Houdini is the TCEC Season 1 Champion.

Congrats!

Final result was 23.5 for Houdini to 16.5 for Rybka.

Best,
Martin