I am sure Vas will do a round up for the tournament, so I'll limit myself to the opening point of view:
Round 1: The Baron - Rybka 0:1
Quite surprisingly The Baron opened 1.e4, which it had never done before against any program that used my book. The reply probably also was a surprise: the French defence. The Baron chose the solid Tarrasch, in which I came up with 3... Be7 and a rather murky line. Rybka quickly equalised and when The Baron refused to exchange queens, she used some nice tactics to win an exchange. Due to the many pawn weaknesses Rybka duly converted the advantage into a win.
Round 2: Rybka - Komodo 1:0
Komodo played a rare line in the Pirc - Ufimzev defence and Rybka quickly grasped the initiative, sacrificing pawns for a swift development. Komodo never got a moment to relax and with some nice blows Rybka won material and the game.
Round 3: Spark - Rybka 0:1
I was a bit surprised to see the O'Kelly Sicilian appearing, which turned out to be a serious problem in my main book. More about that later. Of course white is a bit better in this line and Spark played well to keep some initiative. After a slight mistake Rybka managed to equalize, but Spark refused to trade rooks with a simple draw. Then it went from bad to worse, with Spark playing several sub optimal moves. Rybka outplayed her opponent nicely, ending the first day at 3/3.
Round 4: Rybka - Hiarcs 1:0
Always interesting from a book point of view is the game vs Hiarcs and also this time we were not disappointed. Hiarcs used an offbeat line to combat the 6.h3 Najdorf, but Rybka reacted very strongly by sacrificing a pawn for some serious white square domination. Quietly moving her q-side pawns Rybka got a clear advantage and when the black b-pawn fell, it was the beginning of the end. The strong white b7 pawn decided the game. I liked this game a lot.
Round 5: Shredder - Rybka 0:1
Another surprise awaited me when Shredder opened 1.d4, which was not expected. The book I used for this game had the Dutch Leningrad as a response, one of the offbeat lines I prepared for this tournament. Of course I never thought I'd use it against Shredder :-). White had some edge after the opening, but once again cluster power managed to equalize with some strong moves. For some reason Shredder refused several drawing lines and opted for a dubious looking ending with a weak passed pawn on d6. Another mistake and Rybka nicely rounded up the pawn, converting the endgame into another win.
Round 6: Rybka - Sjeng 0:1
Being tired of so many Ruy's I came up with another surprise for this game, the Scotch. Alas, Sjeng was well prepared and white didn't have anything after the opening. Sjeng could have equalised easily, but preferred to keep the game going. Slowly Rybka was putting her pieces in good positions, showing something like +0.30, when suddenly disaster kicked in. An incorrect exchange sacrifice was played after only a 7 ply search, immediately putting Rybka in a lost position. I was terribly upset and left the tournament hall immediately. My apologies to Gian-Carlo for this behaviour and not congratulate Sjeng on this win, which I'd like to correct here on this forum. After day 2 Rybka was still in the lead with 5/6, Sjeng being 0.5 behind.
Round 7: Junior - Rybka 0:1
To my astonishment I saw Rybka playing the French against Junior, which was totally unintended. First I had the idea Hans had selected the wrong book, but the problem was elsewhere: before round 1 Hans had chosen the wrong book option (I don't know exactly which one) after which a merging of books started. He quickly stopped the process, but apparently some damage was done: the priorities of the book I used against Junior were damaged. Alas, I never would imagine that the book choice actually worked very well. Rybka had simple equality after leaving book and when Junior sacrificed a pawn and refused to go for equality, Rybka quickly took over the game and gave Junior no chance to recover.
Round 8: Rybka - Pandix 1:0
The problem with the book priorities could not be solved before the next round. Against Pandix this led to another Scotch game, which I was not intending. Again, however, the opening went well. Up to 16.h4 it was all book and Rybka quickly got a huge advantage after inaccurate moves by Pandix. A nice win.
Round 9: The King - Rybka 0:1
Finally I managed to catch Hans on the phone trying to do some damage control. Indeed many priorities in the book I wanted to use were altered, so I ordered Hans to change them. Being at home, though, I couldn't check everything and then you know it, you get punished at some point. Another unintended line appeared on the board, leading to a dead equal position after 14 moves. Two times lucky, third time... not. Very frustrating. Luckily The King completely misplayed the position and 10 moves later Rybka was already in control. It didn't take long for Rybka to cash in and hence winning the 10th Leiden ICT with a 8/9 score.
Thanks to everybody for the nice tournament and of course I am happy I can celebrate my return with a Rybka win!
Special thanks to Vas for the great new cluster version; to Lukas for providing this awesome machine; to Hans for operating Rybka this weekend; to Jiri for some nice idea's I could use.
Last but not least: I hope Larry Kaufman will soon recover and can return home.
/* Steinar */
Any idea why She would play the sacrifice after only a 7 ply search?
However, it's kinda sad we don't see some other good engines like Stockfish, Thinker, Bright and Protector, even maybe Crafty
in this tournament.
But it's nice anyway. Congrats Rybka team! :)
- A huge thanks to Lukas for a great cluster and for a lot of last-minute work this week getting everything ready.
- Welcome back, Jeroen, after your long break.
- Thanks to Hans for operating.
- Congrats to Deep Sjeng for the head-to-head win.
- Best wishes to Larry for a fast recovery.
Vas
I'm curious is the Rybka 4 software that run in the cluster the same with Deep Rybka 4 sold? Obviously it's not and it can take advantage of the cluster with an efficient way etc, but i'm asking about evaluation, search techniques, piece values, etc. So?
Vas
>- Congrats to Deep Sjeng for the head-to-head win.
Thanks and same to you for the overall victory. I was hopeful for the tournament victory after our game but then I realized how ruthlessly you guys had ripped through the field whereas I dropped a lot of half points, making it already hopeless to catch up!
This is the second time Deep Sjeng finishes Leiden undefeated and it's not enough for the win.
Our chance to win the tournament given a loss to you can't be more than around 10-15%. The unusually deep field really helped us this time, obviously.
Vas
PS also note that Allard overslept and that it cost Spark a regelmentary loss. This meant that Rybka did not have to face Spark which would have been a tougher opponent as well.
I think Vasik's point was that if the remainder of the field besides Deep Sjeng had been weak (which it was most certainly not!) then the loss in the internal game would have been decisive as everything else would have been sure points. As it was, the rest of the field put up enough of a fight to snatch away some halves from me and give Rybka a chance to win it anyway.
How do you feel about the cluster challenge posted below? I would be happy to organize as the third party in the middle!
> This is the second time Deep Sjeng finishes Leiden undefeated and it's not enough for the win.
Maybe you should set higher contempt when playing against weaker engines
(not Rybka 4).
It is Rybka 5+ AS WELL BEING adapted for use on clusters.
Together, it is the deep blue of 2010.
If i got a Rybka 4 and let it analyze for 100 hours per move, even with as many pc's as is possible, with the deep version, I would never get some of the moves of this Rybka version, as it has extra knowledge which Rybka 4 does not have. It is more than what Rybka 5 will be (even minus the cluster advantage), IF there will be one.
My guess would be that the cluster is different if not say very different from the purchasable Rybka we can have.
We will never have the numbers of a direct match between Cluster Rybka on a 6cores vs Rybka on a 6 cores but I would no be surprised if the cluster is significantly stronger on = hardware
I particularly "like" how naively deceiving "it is based on Rybka 4" sounds
(almost like: it won with 8/9 and it's based on R4 cool, so I have to buy R4)
No doubt late cluster versions (on equal HW) based on 3.xxx numbering were much stronger then public R4 long time before public R4 got released... and then they just changed the numbering.
Probably the correct way to intepret it is that 4.xxx is truly based on Rybka 4
but Rybka 4 Rental which is nothing like Rybka 4 Public
>An incorrect exchange sacrifice was played after only a 7 ply search, immediately putting Rybka in a lost position.
I had a long discussion last year with Vas and Larry about this behavior in R3 Dynamic. I guess Vas kept R3 Dynamic's penchant for exchange sacrifices which could lead rapidly to a lost position as it did in this game. But anyway, congrats to the Rybka team for winning yet another tournament.
but
Round 6: Rybka - Sjeng 0:1
:(
One thing I would note is that the cluster should play quicker with a mate score. At one point it seemed like the cluster had another malfunction when a mate score was shown. A lot of people were quite anxious. Maybe some code that would tell Rybka to play instantly if it sees a mate in less than 20 ... and if during pondering during the opponents time something better comes up, then all the better. It just seems that when there is a mate in say 8, it should play those moves instantly.
>Going through that game there was another blunder that obviously was caused by some malfunction.
Yes, you are right. We were lucky that this bug didn't strike earlier. In a match I played later that day the cluster made several of these blunders in blitz games. It was so bad that we considered going back to the Rybka 3 cluster version we used in Mundial chess. But Vas fixed the bug, I tested the new version and it worked.
>Maybe some code that would tell Rybka to play instantly if it sees a mate in less than 20
That's a very good idea. I'll ask Vas what he thinks about it.
> In a match I played later that day the cluster made several of these blunders in blitz games. It was so bad that we considered going back to the Rybka 3 cluster version we used in Mundial chess. But Vas fixed the bug, I tested the new version and it worked.
Is this bug in the public Rybka 4 version ?
> Maybe we can convince GCP to have a 10 game head to head friendly match on Playchess ... the Clash of the Clusters.
It would be great if somebody could organize something like this. :)
Of course, we could have a "friendly match", but it's always a bit better if there is a third party in the middle.
> One thing I would note is that the cluster should play quicker with a mate score. At one point it seemed like the cluster had another malfunction when a mate score was shown. A lot of people were quite anxious. Maybe some code that would tell Rybka to play instantly if it sees a mate in less than 20 ... and if during pondering during the opponents time something better comes up, then all the better. It just seems that when there is a mate in say 8, it should play those moves instantly.
Yes, of course. I usually add bells and whistles after everything else is good.
Vas
I think the format is not a big problem, but finding someone to organize it in a proper way, just a normal friendly match is a bit boring (I guess you agree :) )
The Cluster play is amazing !
A strong case of "I'm so superior that I can win the event after a depth 7 blunder and running around with a messed up book" :)
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill