Houdini vs Rybka 30 games (up to 24 hours each)

Round 13, Houdini vs Rybka, was a D17: Slav Central, Carlsbad, Main Line, and it ended in a draw.

View game 13 here: http://www.westportchessclub.org/computer-chess/houdini-vs-rybka-30-games-round-13.htm

Rybka leads by a score of 7 - 6

Round 14, Rybka vs Houdini, was a D17: Slav Central, Carlsbad, Main Line, and it was a draw after 126 moves.

Something interesting about this game was the difference in the eval value by both engines. At move 111, Rybka thought it had a +3.34 advantage and Houdini thought Rybka only had a +0.99 advatage (see the print-screen JPG here). And at at move 119, Rybka still thought it had a +3.00 lead, while Houdini thought the game was a draw and valued the position at 0.00. It seems, since I've been running these tournaments (including the 100-game match between these giants), that Houdini is always closer to the correct positional evaluation. The differences in these engines blows me away.

View game 14 here: http://www.westportchessclub.org/computer-chess/houdini-vs-rybka-30-games-round-14.htm

After 14 rounds, Rybka leads by a score of 7.5-6.5

Round 15 is a French Winawer with 7.Qg4 Kf8, and Houdini is up 2.50 pawns at move 31.

Round 15, Houdini vs Rybka, was a C18: French Winawer, 6...Ne7, 7.Qg4 Kf8, and Houdini won.

Five databases (Chess.com, ChessOK.com, 365Chess.com, Rybka 4's Jiri Dufek book that I bought, and Fritz 11's default book) say 7...Kf8 is Black's best chance by a sizable margin, and is the move White has the least success with. I am eager to see how Rybka handles White in this position, but it'll have to wait a week.

View game 15 here: http://www.westportchessclub.org/computer-chess/houdini-vs-rybka-30-games-round-15.htm

After 15 rounds, it's all tied up at 7.5 points each.

so what now , a half time break ? 

thanks for keeping us updated

yes, furtunately and unfortunately, i will be unable to run the chess tournament for about a week. i really want to know how Rybka will handle the White side of the French Winawer!

Game 16 is a French, Winawer: 1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.a3 Bxc3+ 6.bxc3 Ne7 7.Qg4 Kf8 8.h4 Qa5 9.Bd2 Qa4 and is in progress right now. Rybka has White this time, and it played 10.h5 h6. Rybka thought about move 11 for 5,324 seconds: One hour and twenty-eight minutes! It wanted to play 11.Qd1 for the longest time, but at the last second it played 11.Rh3 like Houdini did.

in one screenshot you can peek: its about 42/43 plys

(holy moses...)

In the early stages, it is about 23-26 ply. Near the endgame, it can reach 40+.

Game 16, Rybka vs Houdini, was a French, Winawer, and Rybka won.

I guess the engines proved 7.Qg4 Kf8 is a bad idea for Black, even though it scores the best for Black compared to 7...0-0 or 7...Qc7 in 5 databases. It was interesting to see, however, the engines' different approaches to winning. I really thought I had picked the most even chance by picking 7...Kf8, and when I woke up to see the results this morning, I was a bit deflated. I thought I had chosen a poor opening to have the engines bang their heads on since both proved White's clearly better. So I thought I had sort of let you down by "wasting" two slots of this 30-round competition and thought perhaps I should have chosen 7...0-0 or Qc7 instead. But I think it was an interesting "study" in seeing how each engine went about the win, nonetheless. Houdini did it only 81 moves, and was much more logical about it, where Rybka took 95 moves to mate. Anyway...

Round 16 can be viewed here: http://www.westportchessclub.org/computer-chess/houdini-vs-rybka-30-games-round-16.htm

Rybka leads the match 8.5-7.5

Teary is not so sure that this is a balanced study though, because the opening is not controlled tightly enough.

Engines will usually go out of man made books for around the first 10 moves, so it is not really engine vs. engine at the intial stages. And since humans are prone to innacuracies, then by the time that the engines get out of book and start playing on their own, there could be a slight man-made advantage for one side that would ruin your study (since tenth's of a pawn are a big deal with such high level play).

Possible Solution:

1. Make sure that the engines are using the exact same opening book.
2. After a game has been fully played out (letting the engines choose their openings), go back through it and determine at what point the engines left their books.
3. Start the next game at that same jump off point but with reversed colors, e.g., if Houdini as black got out of book on move 10, then start the next game on move 10 of the previous game with Houdini as white.

Doing that should ensure that the study is perfectly balanced, since human interference will be canceled out and since both sides will have a chance to play from the exact same positions (which will show you the difference in their analysis better),

The point of this 30-game match is to let the computers do the continuations from the most popular human-played openings. The openings I choose are sometimes 9 moves deep and sometimes they are 19. It all depends on where human theory divides for each popular opening. The computer this match is running on is my main PC, so I don't want to let the tournament run a long time. I chose a month, hence 30 games. Since they are playing each opening as White and Black, I get 15 openings. I chose the openings based on feeback from Chess.com forums, members of my chess club, and databases. Again, the point is to play the most popular human-played openings after the current theory runs out or comes to a crossroad.

when you choose the exact SAME openings and just reverse the colours it should be perfectly fair (as long as the engines are not allowed to use any books afterwards).

i find this the most interesting point of your challenge: the computers view of manmade openings. apparently they seem to have refuted one variation already, and given their playing strength being way above that of any human player + way too much time to think it should be clear that this variation is tainted.

Round 17 is a Grunfeld, Exchange, Classical, and is uder way: 1.d4 Nf6 2.c4 g6 3.Nc3 d5 4.cxd5 Nxd5 5.e4 Nxc3 6.bxc3 Bg7 7.Bc4 c5 8.Ne2 Nc6 9.Be3 0-0 10.0-0 Bg4 11.f3 Na5 12.Bd3 cxd4 13.cxd4 Be6 

The 14th move is where the databases seem to come to a cross-roads. Some prefer 14.Rc1 and some prefer 14.d5. I will let Rybka and Houdini figure out what is best for them.

EDIT: Houdini chose 14.Rc1.

Round 17, Houdini vs. Rybka, was a  D89: Grunfeld, Classical Exchange, Main Line, and it ended in a draw. Black was ahead, but could never convert the advantage to a win.

View game 17 here: http://www.westportchessclub.org/computer-chess/houdini-vs-rybka-30-games-round-17.htm

Rybka leads the match 9-8.

Now you can "watch" the Houdini vs Rybka games "live"! To view the current game, go to:

http://www.westportchessclub.org/computer-chess/live-game.htm

I just watched the Kings Indian that was lost by White. It started out very good, but towards the end there were a ton of blunders. Many times pieces were left hanging and not taken. Maybe there was a deeper tactical reason for this, but near the end there were mates in 1 that were missed time and again. I'm only class C, but I can spot a mate in 1. The rooks on both sides were just dancing around pointlessly. As for the mate in 1, it was not forced until White put the rook on b1. Well, maybe it was forced. Either why, Black missed it.

Are these programs running into time trouble at the end of the 24 hours? Maybe you should always give them 60 seconds to make a move, even after 24 hours is exceeded. Fischer style timing would be best.

I looked at the other game, the french where White mated black. That game really got me interested in higher level chess. Black was a piece down, but made what appeared to be an impenetrable fortress. Still, white gradually found a way in without losing any material.

I will have to take a closer look at the game you are talking about. I seriously doubt these engines would have made a blunder so grievous. And the time control is 11 hours and thirty minutes PLUS 30 seconds per move, so the engines will not ever run out of time.