Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / How much is: 'one pawn odd'?
- - By Quapsel (****) Date 2007-07-23 06:35 Edited 2007-07-23 06:42
To get an Idea, what is the worth of 'one pawn odd' I made a pgn-File containing 16 Openings. Alternating one side has one pawn less than the other.
And than I made some matches 'Enginge against a copy of that engine' at several Time controlls.
Under Arena on a PC with Pentium4 with 2,6 MHz and 1GB
The pawn lacks in the games in the order a2, a7, b2, b7, c2...

Rybka 2.3.2a sp
0m+1s     0=10===00000=010   4,5
1m+1s     0=0==00000=00011   4,0
2m+1s     =10==0==000=00=0   4,5
5m+2s     =0000000100=0000   2,0
10m+5s
30m+5s    0000==00=000000=   2,0
100m+5s   =0000000=00=00=0   2,0


TogaII 1.2.1a
0m+1s     =001=0010=1=0==0   6,0
1m+1s     01==110=00=0=000   5,5
2m+1s     001=0=0000000===   3,5
5m+2s     =0000000100=0000   2,0
10m+5s
30m+5s    =00000001=00=000   2,5
100m+5s


Spike 1.2 Turin
0m+1s     =1=0000=0000=0=0   3,5
1m+1s     =1=0=00===0=10==   6,5
2m+1s     ==000=000010===0   4,0
5m+2s
10m+5s
30m+5s
100m+5s


I know, most of these Time Controlls are very short.
But I think, I could see some interesting things:
1.) One pawn odd is a real significant advantage
    Verry short time controlls lead to results near 5:11
    Longer time controlls lead to results near 2:14
    The Programs need a bit more time to ensure the win, after they got a small advantage.
2.) Even with the same Test-Engine a particular missing pawn cann lead to all: win, drawn, loss
    These results were more shuffled than I expected they would be.


During the next days I want to see, what happens, when Rybka has a lack pawn playing against other Enginges
(Toga, Spike, Glaurung...)

Quap
Parent - - By Linus (***) Date 2007-07-23 11:00 Edited 2007-07-23 11:02

>During the next days I want to see, what happens, when Rybka has
>a lack pawn playing against other Enginges (Toga, Spike, Glaurung...)


I can tell you now what is going to happen - Rybka will win. I made a similar test with Rybka vs Deep Fritz 10 at pawn odds, Rybka won 4,5 - 3,5.

http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=4228
Parent - - By Uri Blass (*****) Date 2007-07-23 13:18
I am not sure
Things dependent on the time control and Deep Fritz10 is not the strongest non rybka program and it is possible that the new toga that is going to be released is better at long time control.

From the ccrl list:

7 Deep Fritz 10 4CPU 2925 +22 −22 48.8% +7.3 38.9% 661
69.5%
9 Toga II 1.3 Beta 1 2916 +31 −30 62.9% −88.6 38.9% 352

Uri
Parent - - By George Tsavdaris (****) Date 2007-07-23 13:51

>the new toga that is going to be released is better at long time control.


Is going to be released? When?
That is good news.
The bad news is that i think i have seen that Toga will not be continued by Thomas Gaksch....

Also i saw you saying that you have it and that you believe is not illegal to give it to anyone that wants it.
According to the next you are right.

From the GPL FAQ (http://www.gnu.org/licenses/gpl-faq.html#TOCCanIDemandACopy) :
If I know someone has a copy of a GPL-covered program, can I demand he give me a copy?
No. The GPL
gives him permission to make and redistribute copies of the program if he chooses to do so. He also has the right not to redistribute the program, if that is what he chooses.

That means i can't force you to give it to me, but you have the legal right to give it to me or anyone else if you want.

So if you want, can you please email it to me?
gtsavdar77@gmail.com  (remove 77) is my email.....
Parent - - By Uri Blass (*****) Date 2007-07-23 19:58
I emailed you the version that I got from thomas Gaksch
Parent - By George Tsavdaris (****) Date 2007-07-23 20:01
Thank you.....
Parent - - By sam_i_am (**) Date 2007-07-29 19:01
George ,

f that dude.

makesome code changes of your own . It's a hellofolota fun to see your own MOD beat Rybka . For instance : I had some off days jammed together and simply played with the toga eval and piece values a bit and the attack......oh ya and the move gen( and a few other things have been re-written on the fly just for a simple speed-up).

Anyways , all i'm saying is its dumb to wait on the next version when you have an open code.
Parent - By sam_i_am (**) Date 2007-07-29 19:03
Pawn Odds:

lol........

it's dumb :|
Parent - - By Quapsel (****) Date 2007-07-25 07:02

> I can tell you now what is going to happen - Rybka will win.


I'm surprised.

I've done some Matches Rybka 2.3.2a - TogaII 1.2.1a.
Rybka with one pawn lack in each game.
16 games with each Time control.

The Results for Rybka were:

0m+1s    0=1000=01011==00   6,0
1m+1s    01===10=1000=0=1   7,0
2m+1s    1=1110001=0001==   8,0
3m+1s    1=0==0===10000==   6,0
5m+2s    1=1=1000=1100==0   7,5
20m+5s   10=0=01=00101=1=   7,5
30m+5s   =000100====1000=   5,0
100m+5s


The strength of Rybka shurely ist impressive.
But it seems, that one extra-pawn weights very heavy.
Rybka had more success than TogaII playing against TogaII, but it doesn't suffice to compensate one pawn.
In avarage she got 'only' 42% of the Points.

Now I want to start against Spike.
I would be interested in more Results against newer Versions of Fritz, Hiarcs, Loop, Junior, Shredder...
There should be done more than the 8 Games against Fritz, you have done.
If you are interested, I could EMail you the small PNG-File containing the Openings for pawn-odd, to do some Tournaments.

Quap
Parent - - By Linus (***) Date 2007-07-25 09:55
I can help you with matches against Deep Fritz 10, Hiarcs 11.1 and Zap!Chess Zanzibar. I also own a number of older engines like Shredder 7, Shredder 10 SE, Fritz 5.32 as well as all pusblished Rybka versions (mp + sp, 64 bit + 32 bit).

If you let me know which matches you need I will let them play on my Intel Core 2 Duo 6600, 2.4 GHz, 2 GB RAM, 5 piece + some 6 piece tablebases.
Parent - - By Quapsel (****) Date 2007-07-26 07:46 Edited 2007-07-26 07:54
Hi,

Yes, I think it could be fine to see, how Rybka works with a pawn-lack against a strong new opponent like Fritz, Hiarcs or Zap.
Perhaps one of them with several time controlls. (but I don't want to retard you :-) )
I expect, that these Engines will win more clearly than Toga and Spike, using that pawn-advantage.
But perhaps I will get a surprise.
My experiments are done on a 2,6GHz-P4-Prozessor with Rybka 2.3.2a sp. Shurely against Toga and Spike a mp-Version of Rybka on a dual core Processor will play a little stronger. But would that make so very different Results? Rybka won some 42% against a one-pawn-better TogaII, what could be expected from the MP-Version with a Duo Core Processor?

In these hours I make a Match-Series aganinst Spike, and Spike holds its pawn-advantage similar to Toga, it seems. (I'll post tomorrow)
Next I will take a program, which ist a little weaker (Colossus 2007b oder Aristarch or Anmon or so), to see, whether that strength-Difference equals one-pawn-odd.

Greetings
Quap
quap.sel@gmx.de
Parent - By Linus (***) Date 2007-07-26 15:17 Edited 2007-07-26 15:28
OK, I will run the following tests and post the results here:

Pawn odds matches (Rybka 2.3.2a mp 64 bit with a pawn less) against engines:

1) Hiarcs 11.1
2) Zap!Chess Zanzibar
3) Deep Fritz 10

at time controls:

2m+1s
5m+2s
30m+5s

I will skip the 0m + 1s, 1m+ 1s time controls because I think these are too short, as well as the 10m + 2s (those should not be so different from 5m + 2s) and the 100m + 5s because I don't have that much time.

All matches will be played on one PC (engine match), not computer - computer, on my Intel Core 2 Duo 6600 (2.4 GHz) with 768 MB hash tables for each engine, end game tablebases on (alle 5 plus some 6 piece tables).
Parent - By Vasik Rajlich (Silver) Date 2007-07-26 07:46
This data is interesting - the more the better. According to Larry's investigations, a non-rook pawn seems to be worth 200 to 300 Elo, depending on the time control.

Vas
Parent - By lkaufman (*****) Date 2007-07-23 15:53
     The 2-14 results at long time controls lead me to think we may have a very tough time in the match with Joel Benjamin. At serious time controls, a pawn means much more than it does in blitz.
Parent - - By Quapsel (****) Date 2007-07-27 06:31
Matches Rybka 2.3.2a sp against Spike 1.2 Turin brought the following results.
Rybka won in each set of 16 Games:

0m+1s   10000101111=1=11  10,0
1m+1s   00100001=1==010=   6,0
2m+1s   1=00=1=0=00=1100   6,5
3m+1s   1100100000010011   6,0
5m+2s   000000=001010=10   4,0
20m+5s  =1=011=011=01=1=  10,0
30m+5s  11=0=0010100===1   7,5
100m+5s


Rybka was mostly not able, to reach balance against Spike. But it was close (45%).
(And one can see, that one set of 16 games is much too few, to estimate the relative strength of two engines. The dispersion ist too large. (see the 4,0 and 10,0 in that list. I don't believe, that specially 5 Minutes/game are so fine for Spike and 20 Minutes are so very difficult))

Quap
quap.sel@gmx.de
Parent - - By lkaufman (*****) Date 2007-07-27 15:10
     Just to clarify, is this Rybka 32 bit or Rybka 64 bit?  Thanks for running these tests.
Parent - By Quapsel (****) Date 2007-07-30 07:07
I'm only using Rybka 2.3.2a sp 32 bit.
The PCs available for me for longer Chess-Tests only have a Pentium 4 with 2,6 GHz.
:-(
But I think this is OK to get an Idea of the strength differences of the Engines.

During the last days I made a quick look upon the weaker Engines Colossus 2007b and AnMon 5.60:
Rybka reached here in 16 Games:

against Colossus 2007b
0m+1s   1101=11=01111111   13,0
1m+1s   1==0=11=11101111   12,0
2m+1s   11==1111==111111   14,0
3m+1s
5m+2s   110111=11==10111   12,5


against AnMon 5.60
0m+1s   01111011110=111=   12,0
1m+1s   111010111=111101   12,5
2m+1s    
3m+1s   10=11=1111101111   13,0
5m+2s   1111==11==111111   14,0

The strength-differences are large enough. So Rybka wins very clearly.
To play longer Time Controlls here seems not to be very interesting.

Quap
Parent - By Linus (***) Date 2007-07-27 17:13 Edited 2007-07-27 17:16
Here are my first results from the match Rybka 2.3.2a (mp 64 bit) vs Hiarcs 11.1 at pawn odds, i.e. Rybka had a pawn less in all games, beginning with a2, then a7, then b2, b7, c2 and so on. Rybka had a hard time at short time controls but did much better at longer time controls and managed a 8-8 tie at 30 min + 5 s time controls.

Rybka scores against Hiarcs:
2m + 1s   1010===000000011  5.5 / 16
5m + 2s   ===0=000===0=10=  5.5 / 16
30m + 5s  1000=01110001=11  8.0 / 16


That is impressive, to say the least. Hiarcs 11.1 is listed as the #3 engine at CCRL with about 130 Elo less than Rybka.

Next match: Zap!Chess Zanzibar, probably finished by Sunday.
Parent - - By Linus (***) Date 2007-07-28 15:54
Here are my results from the match Rybka 2.3.2a vs Zap!Chess Zanzibar at pawn odds, i.e. Rybka had a pawn less in all games, beginning with a2, then a7, then b2, b7, c2 and so on. Rybka won at all time controls.

Rybka scores against Zap!Chess Zanzibar:
2m + 1s   1=0===011===0111  9.5 / 16
5m + 2s   =11==11=0==01=10  9.5 / 16
30m + 5s  1====0110======1  9.0 / 16
Parent - - By lkaufman (*****) Date 2007-07-29 13:29
Wow, this result is amazing! Any idea why Zap did so much worse than Hiarcs?
Parent - - By Linus (***) Date 2007-07-29 18:40
No. I would have expected more from the #2 program.
Parent - - By Vasik Rajlich (Silver) Date 2007-07-30 10:07
This result may not really be anomalous. I suspect that on your 32-bit machine, Hiarcs is stronger than Zap.

Vas
Parent - By Linus (***) Date 2007-07-30 11:28
Actually these tests were done on a Dual Core computer with 64 bit Vista OS and Rybka 2.3.2a mp 64 bit. I do not know if Hiarcs 11.1 is a 64 bit software (but it did use both processor cores). Zap!Chess Zanzibar definately was only the 32 bit version (but also a mp version). The Zap!Chess Zanzibar 64 bit version does not run in the version 10 Chessbase GUI.
Parent - By Quapsel (****) Date 2007-07-27 07:20
You can find the pgn-File I'm using with the pawn-odd-Openings under
http://de.geocities.com/quap.selchen/bauernvorteil.txt

Quap
Parent - - By lkaufman (*****) Date 2007-07-30 19:00
     Here are some Elo values based on your self-play tests. The average Elo value for all time controls for a pawn was 209. For White pawns only, it was 180. For Black pawns only, it was 230 (these numbers may be off a couple points due to rounding in the FIDE calculator I used). For edge pawns, it was 133. For non-edge pawns, it was 235. This seems completely consistent with my earlier findings that a non-edge pawn is worth about 200 in blitz and about 300 in slow chess.
     Since these numbers averaged all time controls, they are dominated by the blitz results. The average for the longer time controls (30' and up) was an Elo value of 309 for a pawn. Based on this, the equivalent rating of Joel Benjamin at this handicap would be just a tad under 2900. 
Parent - - By Linus (***) Date 2007-07-30 19:21
Assuming that Rybka's rating is around 3000, this would mean that Rybka is still the favourite. Then again, considering Joel's experience in playing against computers, it could be a close call. This should guarantee interesting games after all.
Parent - - By lkaufman (*****) Date 2007-07-31 03:57
Another factor in Joel's favor is that the handicap completely kills the computer's opening book (although I will create a tiny one for the match), whereas Joel can still play standard openings "by analogy", something which today's chess programs don't do. This might be worth another fifty Elo or so.
Parent - By Vasik Rajlich (Silver) Date 2007-08-01 07:33
FWIW - I voted 4.5-3.5 in favor of Rybka. At even odds, I'd still bet on her.

Also, by the way, my predictions were exactly right in the two matches vs Ehlvest. Obviously, there is a lot of luck in an 8-game match, but in both previous matches, the paper-napkin Elo calculations turned out to be accurate.

Vas
Parent - - By Quapsel (****) Date 2007-08-02 12:08
At least I want to post the Results of Rybka 2.3.2a sp 32bit with one absent pwan against Fruit 2.3

0/1    10100000=1=01000   5.0
1/1    1010000=000=001=   4,5
2/1    10===000001==0=0   5,0
3/1    1=0000=110=00=0=   5,5
5/1    0=0100=0=1100110   6,5
10/3
20/3
30/5   ===000==0==0=101   6,0


The results are rather similar to those against Toga.

And what can Rybka do wit such an advantage?
I rerun that match, but now Fruit had to handle with one absent pawn.
Rybka won:

0/1   0111=10=1==11110   11,0
1/1   111==0=11=1111=1   12,5
2/1   111=1=1111111101   14,0
3/1
5/1
10/3
20/3
30/5  =111=11111=01111   13,5


Here the better playing stregth of Rybka can be seen again.
But it would be estimated too much, if i would assume, that
Rybka does not give away any point in that situation.

Quap
Parent - - By SR (****) Date 2007-08-02 12:30
A techincal question: How do I play pawn handicap match involving two different programs under the Fritz 9 GUI?
Parent - By Linus (***) Date 2007-08-02 15:23
You store the position(s) with a pawn less in a database. Then you start an engine - engine match using those positions from the database.
Up Topic Rybka Support & Discussion / Rybka Discussion / How much is: 'one pawn odd'?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill