Houdini 3 vs Cluster Rybka Game 6 ...

[Uri Blass]

who care if the cluster could lose on time?

I think that most people were going to consider the result as 4.5:1.5 for houdini even in case that the cluster lost on time.

The fact that technically the result is 5.5:0.5 for houdini in that case is clearly uninteresting because the idea was not to test the cluster for tournaments
but to test the quality of moves that the cluster suggest to know if it is useful for analysis.

In positions of simple forced mate or a very big advantage nobody even need to use the cluster and the cluster can be clearly replaced by rybka or stockfish or critter on normal hardware that does not lose on time so I think that using the cluster is a waste of money for no benefit.

[IWB]

the technical question if the cluster can handle such a time problem and still win (or just draw or even lose) without the right 6pc tbs is interesting.

It has done so before in blitz tournaments. And how is an engine in time trouble with a 15 second per move increment when there are only about 15 moves left ?

If you play a blitz tourney I suspect that your whole setup is different that in a long game ... and looking at the time consumtion in the last moves the setup doesn't look well suited for blitz ...

Anyhow, we will never know.

Bye
Ingo

[IWB]

Nonetheless as I could see and was informed that the setup of the cluster was differrent in any game I would be interested if ...

But you are right, for the overall result and a position with a forced mate is really does't matter.

BYe
Inog

[beram]

The theme of the weak bishop, with pawns of the same color was significant in this game. I have stated earlier (http://www.talkchess.com/forum/viewtopi ... 52&t=45776) that this could be a slight weakness of Houdini 3, which could be improved by Robert Houdart.
The white bishop was weaker than the black bishop and on the other hand white could not benefit of the weak pawn on c6.
Nice game to watch and strong play by the Rybka cluster.

grts Bram

[M ANSARI]

Obviously a thing you cant make right as an organizer.

If we adjucate 15s before the end it was wrong as it either might be a loss or because we adjucated at all that short before the end to "let it look like ..."

If we woud not have adjucated the game and it would have lost it would be foul play as the cluster had a disadvantage.

In conrtrary to you I am less sure that the shown mate would be correctly executed. And actually I would be a bit interested to see if it would in time but decided that this woud destroy the nice game of Rybka before.

Check this:
88. Ra7 {-13.80/22 9 (f4)} Rc8 {-#18/24 351 (Te3)}
89. Kb1 {-13.79/28 24} Rb8+ {-#16/26 171 (Th8)} 90. Kc1 {-10.99/26 23 (Ka1)}
Kd4 {-#17/24 182 (Kb4)} 91. Rd7+ {-12.22/29 73} Ke3 {-#16/26 0} 92. Re7 {-12.
22/25 7} Rc8+ {-#15/29 83 (Kxf3)} 93. Kb1 {-19.95/27 10} Kxf3 {-#15/29 75} 94.
Ng5+ {-#22/35 0} Kf4 {-77.98/20 1} 95. Ne6+ {-#19/34 20} Kf5 {-#13/28 61} 96.
Ng7+ {-#19/34 23} Kg6 {-#13/29 55} 97. Re6+ {-#17/33 0 (Te8)} Kxg7 {-#12/32 135
} 98. Re2 {-#15/39 0} Kg6 {-#11/32 94 (Kf6)} 99. Rf2 {-#15/38 89 (Th2)} Kg5 {
-#11/37 108 (Tcd8)} 100. Ka1 {-#14/36 28 (Ka2)} Kh4 {-#10/39 69 (Kg4)} 101. Rg2
{-#13/36 6 (Te2) Felix 2,Houdini 3 Pro x64 gibt auf (Lag: Av=0.24s, max=0.8s)}
0-1

Starting move 88 it shows a mate in
18, 16, 17, 16, 15, 15, Eval:-77.98, 13, 13, 12, 11, 11, 10, 13 - so the Cluster was going up and down ... and the correct mate at move 98 is a #15. When Rybka is Showing a mate in 10 it is a mate in 13. Most moves in between the the first 6pc and the last move were not optimal!

Then the times per move in seconds starting at 88 (the first mate score):
351, 171, 23, 182, 0, 83, 75, 1, 61, 55, 135, 94, 108, 69.

All this with 15s left on the clock at the end (and you can calculate how much was left prior to the last 2 moves for example.

Let your prejudges about why this games was adjucated aside, the technical question if the cluster can handle such a time problem and still win (or just draw or even lose) without the right 6pc tbs is interesting. Unfortunately to answer this question it would have destroyed the very nice strangualting before ...

So we are coming back to: "Whatever we do it is wrong!"

Actually now I wish we haven't seen the problem and just would have let it run until the (bitter) end.

Bye
Ingo

Maybe that was the cluster's way of milking maximum ducats possible Actually that could also point out to a general problem with the cluster in game play as it did play several suboptimal moves in its matches that were hard to understand. At this level one small misstep is all it takes to lose. Maybe there is a problem between the motherboards in communication. I do know that in my Rybka 4.1, I will get a stall every once in a while where the kns simply drop to almost nothing. I can see a problem where if the engine on one of the slave motherboards stalls at a higher evaluation, the master computer will not know it and if the evaluation is higher, it will choose that move thinking that the evaluation came about after deep search.

[IWB]

Maybe that was the cluster's way of milking maximum ducats possible ..

That is a creative thought indeed!

Btw: I have to thank Lucas Cimiotti as well. He helped a lot with the cluster and gave usefull informations for the games.

Thanks Lucas!

Regards
Ingo

[beram]

12...b5! appears to be a very good move by Rybka, unexpected for Houdini and equalizing completely.

Yes and last move 17 ..a5 ! also

After further analysing with Deep Rybka and Houdini I come to conclusion that after move 16 Bf4 ?! white comes behind black
[d] r4rk1/2nqbp1p/2p1bpp1/p7/2BP4/P4N2/1P1B1PPP/2RQR1K1 w - - 0 18
After that move black responds with …Bd6 ! and after the intermediate exchange of white fielded bishops on e6 white must retreat to a passive place on e3, whereafter the black bishop is better than white one, and whites counterplay is slightly diminished to nothing.

Giving Houdini enough time he prefers Bxe6 and Qc2 at depth 26
with this lines: Analysis by Houdini 3 x64 26ply:

1. = (0.00): 18.Bxe6 Nxe6 19.Qa4 Rfc8 20.Rc3 Rab8 21.Rc2 Bd8 22.d5 Qxd5 23.Bc3 Rb5 24.Rd2 Qf5 25.Nh4 Qh5 26.Nf3 Rd5 27.g4 Qh3 28.Rxd5 cxd5 29.Rxe6 Rc4 30.Qe8+ Kg7 31.Ne5 Rxg4+ 32.Nxg4 Qxg4+ 33.Kf1 Qh3+ 34.Kg1 Qg4+ 35.Kf1
2. = (0.00): 18.Qc2 Nd5 19.Bh6 Rfb8 20.Ba2 Ra6 21.Nd2 a4 22.Nc4 Nc7 23.Be3 Nd5 24.Bh6 Nc7

Deep Rybka in compare analysis prefers Qc2 giving -0,02 at depth 20
I checked the lines and run it through with both engines and so came to the above mentioned conclusion

"just my few cents" giving food for thought
grts Bram

[beram]

My final quote on this matter: 34. Qe2 was the final mistake and loosing move by Houdini. 34. Re3 (Re2 or Re1) would have saved the game
After 34. Qe2 ? - h6 ! the rook on e-file gets trapped as in the game or white looses pawn d4 when he wants to escape from the loss of exchange.

[d] 1b6/5pkp/1rp1npp1/pr1q4/2RPR2P/P4N2/1PQ2PP1/2B3K1 w - - 0 1

Analysis by Houdini 3 x64 32 ply (>2 hours):

1. =/+ (-0.38): 34.Re3 Bf4 35.Rxe6 Bxc1 36.Rexc6 Rxc6 37.Rxc6 Bf4 38.Rc8 Qa2 39.Ne1 Qxb2 40.Qxb2 Rxb2 41.Kf1 Bd6 42.Ra8 Bc7 43.Nf3 Ra2 44.Ra7 Bb6 45.Ra6 Bd8 46.Ra8 Bc7 47.Ra7
2. =/+ (-0.38): 34.Re1 Bf4 35.Rxe6 Bxc1 36.Rexc6 Rxc6 37.Rxc6 Bf4 38.Rc8 Qa2 39.Ne1 Qxb2 40.Qxb2 Rxb2 41.Kf1 Bd6 42.Ra8 Bc7 43.Nf3 Ra2 44.Ra7 Bb6 45.Ra6 Bd8 46.Ra8 Bc7 47.Ra7
3. =/+ (-0.38): 34.Re2 Bf4 35.Rxe6 Bxc1 36.Rexc6 Rxc6 37.Rxc6 Bf4 38.Rc8 Qa2 39.Ne1 Qxb2 40.Qxb2 Rxb2 41.Kf1 Bd6 42.Ra8 Bc7 43.Nf3 Ra2 44.Ra7 Bb6 45.Ra6 Bd8 46.Ra8 Bc7 47.Ra7

(Bram, Sommelsdijk 21.12.2012)