PDA

View Full Version : Rybka



pax
14-12-2005, 06:31 AM
Did anyone get hold of the Rybka beta download before it was removed? Apparently it has tested very strongly vs Fritz, Shredder and Fruit.

pax
14-12-2005, 06:41 AM
Actually, I just found it on eMule.

pax
19-12-2005, 08:54 AM
I've been doing some testing in Arena between Rybka, Toga, Fruit and Crafty. Rybka kills them all. In a 16 game match with Toga (the strongest of the other 3), Rybka scored +7 -3 =6.

What is particularly remarkable, is that Rybka achieves these results with a node count of about 10% of the other engines. In other words, it looks at 1/10th as many positions per second as other engines. While everyone else is streamlining their evaluation functions to enable more positions per second, Rybka is using *more* knowledge, looking at fewer positions, and heavily pruning the search tree.

Davidflude
22-12-2005, 03:56 PM
I have just installed fruit. It was a real joy to install. I am running it in the Shredder interface. I shall download the other program later. At the moment I have the two Box Hill Chess Club computers at home (storing them during our move). I shall test out these programs using ome of my completed correspondence games and compare their results with those of Tim Harding in Chess Mail. He missed a win for my opponent but my opponent missed it as well.
(I saw it but as I could not prevent it ignored it. )

pax
22-12-2005, 04:27 PM
If you're using the free version of Fruit (rather than the commercial 2.2.1), then you should also get hold of Toga II. Toga is based on the free version of Fruit, but is (quite a bit) stronger. It beats the latest versions of Shredder, Junior and Fruit 2.2.1, and does better than any other engine against Rybka (but Rybka is king for now).

Also if you're looking for a great UCI/Winboard interface, check out Arena. It's very easy to use, and especially to set up multi-engine tournaments.

Davidflude
23-12-2005, 09:41 AM
If you're using the free version of Fruit (rather than the commercial 2.2.1), then you should also get hold of Toga II. Toga is based on the free version of Fruit, but is (quite a bit) stronger. It beats the latest versions of Shredder, Junior and Fruit 2.2.1, and does better than any other engine against Rybka (but Rybka is king for now).

I am thinking of buying the full version of Rybka and possibly fruit. Is Toga II stronger than the commercial version of Fruit?




Also if you're looking for a great UCI/Winboard interface, check out Arena. It's very easy to use, and especially to set up multi-engine tournaments.[/QUOTE]

I shall not be using UCI/WINBOARD I shall evaluate the programs using completed correspondence games. For example I will use Estrin/Berliner which
has been analysed in enormous depth.

pax
23-12-2005, 01:41 PM
I am thinking of buying the full version of Rybka and possibly fruit. Is Toga II stronger than the commercial version of Fruit?


Toga is the strongest free program by a way. It is reported to be close to commercial Fruit in strength. Check out the 5moves rating list:
http://www.utzingerk.com/rating_5moves.htm

(neither Toga nor Rybka are on the more official SSDF list yet)

Gringo
26-01-2006, 08:35 AM
the loss by Rybka against Spike was surprising, considering the other impressive wins.:eek:

Davidflude
26-01-2006, 09:42 AM
I am going to talk a little about testing the three above mentioned programs as analysis machines.

I am going to use a variation which was originally published in Kaissiber magazine but where I have found an improvement

The test position arises as follows:

1.d4 d5
2.e5 dxe
3.Nc3 Nf6
4.f3 exf
5.Nf3 Bf5
6.Bc4 e6
7.0-0 c6
8.Ne5 Bxc2
9.Nxf7 Kxf7
10. e2 Qxd4
11. Be3 Qxe3
12. Kh1 Bd6

Initially all programs assess this program as very good for black. If you leave the programs running for long enough they will eventually discover that white has at least a draw with best play. How long depends on your machine configuration and the size of your hash tables. I would expect any good computer program to find the moves but the question is how long would it take to do so.

I have seen suggestions that all you need to do in correspondence chess is let your computer program analyse the position for several days. There is a wee problem. If you are playing a reasonable number of games, say 15-20 you would lose all the games on time.I shall post some of the analysis in a few days. Also I intend to post a variation from a completed game where I suspect that the computer programs will not have a clue. Nor would most human beings.

Rincewind
26-01-2006, 10:37 AM
There is a wee problem. If you are playing a reasonable number of games, say 15-20 you would lose all the games on time.

Easy solution?

number of computers = number of games

Vlad
26-01-2006, 10:59 AM
Easy solution?

number of computers = number of games

Even easier - just connect to ANU's super computer. The task will be done in no time.

sharpnova
10-03-2006, 06:41 AM
2r1qrk1/pb1n1pp1/1p2pn1p/6B1/2BP4/1QR5/PP1N1PPP/4R1K1 w - - 0 16

i was wondering if anyone who had the latest version of rybka could test this position and see how long it takes rybka to find 16. Bxh6

pax
10-03-2006, 02:54 PM
2r1qrk1/pb1n1pp1/1p2pn1p/6B1/2BP4/1QR5/PP1N1PPP/4R1K1 w - - 0 16

i was wondering if anyone who had the latest version of rybka could test this position and see how long it takes rybka to find 16. Bxh6

It took about 8-9 minutes on my fairly sluggish P4. Bear in mind that Rybka is a 'positional' engine, and doesn't look at nearly as many nodes as other engines. Other engines may well be quicker.
Edit: make that less than 7 minutes..

Edit: Also note that this is the free beta 1.0 version. Not that I would expect the latest version to be any different.

pax
10-03-2006, 03:07 PM
Actually Toga took a bit longer (about 10 minutes) looking at 10x as many nodes.

lilblam
12-03-2006, 04:08 AM
2r1qrk1/pb1n1pp1/1p2pn1p/6B1/2BP4/1QR5/PP1N1PPP/4R1K1 w - - 0 16

i was wondering if anyone who had the latest version of rybka could test this position and see how long it takes rybka to find 16. Bxh6

On AMD Athlon 64 3500+ (2.2Ghz) using Rybka 1.01 Beta 13d 32-bit, it took 1 min, 47 seconds:

Rybka 1.01 Beta 13d 32-bit:
3 00:00 320 20.480 +0.15 Bg5f4
3 00:00 640 20.480 +0.21 Bg5h4
4 00:00 1.760 22.813 +0.22 Bg5h4
5 00:00 4.672 25.447 +0.32 Bg5h4 Nf6d5
6 00:00 8.768 28.685 +0.25 Bg5h4 Nf6d5 Rc3g3
7 00:00 17.584 38.392 +0.25 Bg5h4 Nf6d5 Rc3g3 Nd7f6
8 00:01 38.232 47.225 +0.20 Bg5h4 Nf6d5 Rc3c2 Nd7f6 f2f3
8 00:01 48.968 51.747 +0.31 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a3 Nd7f6
9 00:01 81.328 57.276 +0.39 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a3 Rc8xc3 b2xc3 Qe8a8
10 00:02 135.760 61.758 +0.29 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a3 Qe8d8 Rc3xc8 Qd8xc8 Qa3xa7
11 00:04 278.888 66.693 +0.18 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3c2 Rc8xc3 b2xc3 Qe8a8 Nd2e4 Bd5xe4
11 00:13 809.832 64.404 +0.40 Bg5h4 Nf6d5 Rc3g3 Nd5f4 Nd2e4 Bb7xe4 Re1xe4 Nf4d5 Re4g4
12 00:22 1.460.080 66.402 +0.38 Bg5h4 Nf6d5 Rc3g3 Nd5f4 Nd2e4 Bb7xe4 Re1xe4 Nf4h5 Rg3c3 g7g5
13 00:36 2.309.192 64.755 +0.34 Bg5h4 Nf6d5 Rc3g3 Nd5f4 Nd2e4 Bb7xe4 Re1xe4 Nf4h5 Rg3c3 g7g5 Bh4g3
14 01:04 4.270.832 67.753 +0.40 Bg5h4 Nf6d5 Rc3g3 Nd5f4 Nd2e4 Bb7xe4 Re1xe4 Nf4h5 Rg3c3 g7g5 Bh4g3 Nh5xg3
14 01:47 6.868.232 65.748 +0.65 Bg5xh6 g7xh6 Rc3h3 Nf6h7 Qb3e3 Nh7g5 Rh3xh6 Qe8d8 Rh6h5 f7f6 Bc4xe6+ Kg8g7

---------------

On the same machine, it took Toga II 1.1a exactly 6 mins 17 seconds.

Toga II 1.1a:
1/2 00:00 4 0 +0.33 Bg5xf6 Nd7xf6
1/9 00:00 353 0 +0.36 Bg5e3
1/9 00:00 354 0 +0.78 Bg5f4
2/12 00:00 745 0 +0.90 Bg5f4 Nf6g4
3/12 00:00 1.787 0 +0.62 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5
4/12 00:00 2.961 0 +0.55 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Nd2c4
5/14 00:00 7.307 0 +0.60 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Nd2c4 Bd5xc4 Rc3xc4
6/21 00:00 20.502 0 +0.49 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3c2 Rc8xc3 b2xc3 Qe8e7
7/21 00:00 57.121 0 +0.54 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a3 a7a5 Bf4c7 h6h5
8/21 00:00 127.253 0 +0.55 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a3 a7a5 Nd2e4 Bd5xe4 Re1xe4 Rc8xc3 b2xc3
9/23 00:00 311.999 0 +0.67 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3a4 a7a5 Bf4d6 Rc8xc3 b2xc3 Nd7f6 Qa4c2
10/25 00:01 609.935 0 +0.49 Bg5f4 Nf6d5 Bc4xd5 Bb7xd5 Qb3c2 f7f5 Bf4d6 Qe8g6 f2f3 Rc8xc3 b2xc3 Rf8c8
10/25 00:02 2.056.883 785.000 +0.66 Bg5h4 Nf6d5 Nd2e4 Nd5xc3 Ne4d6 Rc8xc4 Nd6xe8 Nc3e2+ Kg1f1 Rc4xd4 Re1xe2 Rf8xe8
11/28 00:04 3.518.907 835.000 +0.61 Bg5h4 Qe8e7 Bc4b5 Rf8d8 Qb3a4 a7a5 Bb5xd7 Rc8xc3 Bh4xf6 Qe7xd7 Qa4xd7 Rd8xd7 b2xc3 g7xf6
12/32 00:09 7.647.012 848.889 +0.56 Bg5h4 Qe8e7 Bc4b5 Rf8d8 Qb3a4 a7a5 Bh4g3 Rc8xc3 b2xc3 Nf6g4 Bb5d3 Nd7c5 d4xc5 Rd8xd3 Qa4xg4 Rd3xd2 c5xb6 Rd2xa2
13/32 00:15 13.477.786 870.000 +0.53 Bg5h4 Qe8e7 Bc4b5 Rf8d8 Qb3a4 a7a5 f2f3 Nd7f8 Rc3xc8 Rd8xc8 Nd2c4 Nf8d7 Bh4xf6 Nd7xf6 Nc4xb6
14/38 00:35 30.792.410 886.176 +0.59 Bg5h4 Qe8e7 Bc4b5 Rf8d8 Qb3a4 a7a5 f2f3 Rc8xc3 b2xc3 g7g5 Bh4f2 Nf6d5 Qa4b3 Nd7f6 Bf2g3
15/42 01:33 83.601.067 897.419 +0.34 Bg5h4 Qe8e7 Bc4b5 Rf8d8 Qb3a4 Rc8xc3 b2xc3 a7a6 Bb5xd7 g7g5 Bd7c6 b6b5 Qa4a5 Bb7xc6 Qa5xa6 Qe7c7
16/64 04:00 214.591.348 894.754 +0.35 Bg5h4 Qe8e7 Nd2f3 Qe7d6 Bh4g3 Qd6e7 Rc3e3 Nf6h5 Bg3h4 Nh5f6 d4d5 Qe7c5 d5xe6 Qc5xc4 e6xd7 Nf6xd7 Qb3d3 Qc4xd3 Re3xd3
16/64+ 06:17 336.703.421 892.405 +0.74 Bg5xh6 g7xh6 Rc3h3 Nf6h7 Rh3xh6 Kg8g7 Rh6xh7+ Kg7xh7 Qb3h3+ Kh7g7 Qh3g3+ Kg7h8 Qg3h4+ Kh8g7 Qh4g5+ Kg7h7 Re1e3 Bb7f3 Re3xf3 f7f5 Rf3g3 Kh7h8 f2f3

Thunderstone
31-05-2006, 07:51 AM
Rybka is coming equal third in the current World Computer Championship.
It's playing under the name "Rajlich".

http://www.cs.unimaas.nl/wccc2006/index.htm

Note this event is being held along side the Olympiad in Turino.

Davidflude
02-06-2006, 11:31 AM
I am comparing the free version of Rybka with Shredder8. On my configuration
Shredder is clearly faster but rybka is nevertheless very strong.


I have a dual core athlon 3800 with decent graphics card and 2 megs of memory. I am curently looking at one of Zhao's games from the Olympiad. Of course he had to find a move in minutes wheras I can let the computer run over night.

pax
02-06-2006, 01:52 PM
I am comparing the free version of Rybka with Shredder8. On my configuration
Shredder is clearly faster but rybka is nevertheless very strong.


How do you mean Shredder is "faster"?

If you measure nodes per second, any other engine comes out an order of magnitude quicker than Rybka. The remarkable thing is that Rybka is generally stronger even though it looks that a fraction of the positions.

Garvinator
02-06-2006, 04:00 PM
How do you mean Shredder is "faster"?

If you measure nodes per second, any other engine comes out an order of magnitude quicker than Rybka. The remarkable thing is that Rybka is generally stronger even though it looks that a fraction of the positions.
Thought you might be interested that Junior has just won the latest world computer chess championship held in conjunction with the Olympiad.

http://www.chessbase.com/newsdetail.asp?newsid=3153

pax
02-06-2006, 04:20 PM
Thought you might be interested that Junior has just won the latest world computer chess championship held in conjunction with the Olympiad.

http://www.chessbase.com/newsdetail.asp?newsid=3153

Yep, so I see.

It is, of course, too few games to make a strong judgment about the relative strength of the engines.

Garvinator
02-06-2006, 04:21 PM
Yep, so I see.

It is, of course, too few games to make a strong judgment about the relative strength of the engines.
yeah i know and also in the computer world championship, alot depends on how the programs were set up by the designers, specifically for the tournament ie opening prep for the other engines.

So for general run of the mill play, it is only a guide.

pax
02-06-2006, 05:52 PM
The 5 moves rating list is now here:

http://www.utzingerk.com/rating_list.htm

This is a good bet for 'standard' versions of each product, and standard computer hardware.

The 1.2 version of Rybka absolutely creamed Shredder 10 in a 50 game match (+27, -8). Makes Rybka's loss to Shredder in the WCCC burn a bit!

Davidflude
03-06-2006, 12:54 AM
How do you mean Shredder is "faster"?

If you measure nodes per second, any other engine comes out an order of magnitude quicker than Rybka. The remarkable thing is that Rybka is generally stronger even though it looks that a fraction of the positions.

I agree I set up some fearsomely complex positions where the evaluation of the position changes as the program analyses deeper in the position. Shrdder took less elapsed time for the evaluation to change. Of course this in no way tests how the programs compare in more normal positions.

pax
05-06-2006, 11:03 AM
I agree I set up some fearsomely complex positions where the evaluation of the position changes as the program analyses deeper in the position. Shrdder took less elapsed time for the evaluation to change. Of course this in no way tests how the programs compare in more normal positions.

Fair enough.

That's probably the sort of test that Rybka will not be as good at. It performs well by making more sophisticated evaluation on a shorter search, rather than by looking very very deep.

Thunderstone
14-06-2006, 11:32 AM
The new MP version of Rybka (just released) is already showing promise :)

http://www.husvankempen.de/nunn/40_4_Ratinglist/40_4_AllVersion/rangliste.html