Stockfish 2.1 running for the IPON

[IWB]

as usual:

http://www.inwoba.de

Have fun
Ingo

[Vinvin]

Code: Select all

309.5	-	107.5		74.22%		Perf=2949

Currently +29 points over 2.01, great !

[IWB]

Code: Select all

309.5	-	107.5		74.22%		Perf=2949

Currently +29 points over 2.01, great !

Be carefull,

1. It is only a bit over 400 games and
2. The calculation is more Elostat than Bayes. The final result (with Bayes) will be a few Elos lower.


Bye
Ingo

[Leto]

Code: Select all

309.5	-	107.5		74.22%		Perf=2949

Currently +29 points over 2.01, great !

Be carefull,

1. It is only a bit over 400 games and
2. The calculation is more Elostat than Bayes. The final result (with Bayes) will be a few Elos lower.


Bye
Ingo

Now after 770 games Stockfish 2.1 is at 2941, 21 elo higher than Stockfish 2.01. It is looking like there is a measurable strength improvement in Stockfish 2.1.

[Leto]

After 918 games it is down to 2929, only 8 elo higher, probably no elo increase or very little.

[Frank Quisinsky]

Please thinking on the remis quote.

Very easy ...
With a highter remis quote = - 7-11 ELO from Shredder tourney table calculation to bayesian calculation.

Means:
If Stockfish 2.1.0 have the same remis quote as in SWCR (38%) you can calculate - 7-11 ELO.

IPON Test = 2.932 - 11 = 2.921 ELO (1 ELO more as version 2.0.1).

If Stockfish have 33% remis quote (same as version 1.8.0 or 1.9.1) you have to calculate around -5 to Bayesian.

We don't know the remis quote from Stockfish 2.1.0 x64

But now after around 1.000 games up to 2.000 games SF will get maximal +-5 and with a high remis quote the result will be max. + 0-10 to 2.0.1!

ELO-LaOla
From game number 500 - 1000 the ELO-LaOla is 10
From game number 1.000 - 1.500 the ELO-LaOla is 5
From game number 1.500 - 2.500 the ELO-LaOla is 3-4

This one is the SWCR ELO-LaOla if you have around 25 opponents. A little bigger LaOla you get if you have 20 opponents!

Allways the same in 140 of 143 cases (tested SWCR engines). Only in three cases the ELO-LaOla is 21, 21, 22 points from game number 500-1000.

Interesting is the w32 version.
Stockfish 1.9.1 w32 is 20 ELO stronger as Stockfish 2.0.1. If the new version 2.1.0 is 10 ELO stronger only, the best w32 version will be again 1.9.1. The most interesting chess after analyzes (you can do it too with SWCR database, all games are available with the double time as IPON) played version 1.7.1!

Best
Frank

[Graham Banks]

Interesting is the w32 version.
Stockfish 1.9.1 w32 is 20 ELO stronger as Stockfish 2.0.1. If the new version 2.1.0 is 10 ELO stronger only, the best w32 version will be again 1.9.1. The most interesting chess after analyzes (you can do it too with SWCR database, all games are available with the double time as IPON) played version 1.7.1!

Best
Frank

In my CCRL 40/40 32-bit 1CPU testing, I have Stockfish 2.0.1 as about 15 ELO stronger than the previous version. Can't say anything about the latest version yet. though.
http://computerchess.org.uk/ccrl/4040.l ... ons_only=1

Cheers,
Graham.

[Frank Quisinsky]

Hi Graham,

around the same number of games (CCRL to SWCR), between version 1.9.1 and 2.0.1. In this case we should search a bit deeper.

How many opponents you have in CCRL?
In SWCR very easy ... games : 40 = opponents because each match since SWCR game number 1 = 40 games.

Or ...
How many bogey opponents or wish opponents. We have in computer chess the problem, that many interesting parts of sources are cloned. Good example is LMR / Null-Move.

Could be 10-20 ELO ... in a rating list.

The biggest problem for us, means people with working on a rating list. So it's not important how many games you have, more important is how many opponents you have and have the opponents the same strengths or weaknesses.

And ...
We can nothing do against it.
Ratings can be unclear.

I believe that all ratings from the TOP-20 are unclear to 10-20 ELO. Not important how many games you have.

A good example is ChessTiger.
No changes since many years.

Now, test ChessTiger vs. actual engines. ChessTiger have different strenghts, each time a remis is possible, vs. 500 ELO more too. But for a modern engine not possible. Could be 10 ELO! Means, that ChessTiger will be play today around 10 ELO stronger if you test vs. modern / new engines. If you test vs. engine, in the time ChessTiger was available ... ChessTiger will have 10 ELO fewer.

That I mean with ... a lot of strong sources are copy with little changes in modern engines, more knowledge today is available.

Since Fruit!
With the idea to create a source code more easy ... not complicated with more chess knowledge. Other follow the main idea and more and more engines have the same playing style.

Best
Frank

But all in all ...
No clear improvments in from Stockfish 1.7.1 to 1.8.0 to 1.9.0 to 2.0.1. In this case more interesting as ELO is the playing style and after all anylzes I can make its for me clear that the most interesting chess plays version 1.7.1.

[Graham Banks]

Hi Graham,

around the same number of games (CCRL to SWCR), between version 1.9.1 and 2.0.1. In this case we should search a bit deeper.

How many opponents you have in CCRL?
In SWCR very easy ... games : 40 = opponents because each match since SWCR game number 1 = 40 games.

http://computerchess.org.uk/ccrl/4040.l ... 0_1_32-bit

[Frank Quisinsky]

Hi Graham,

could you send me all the CCRL Stockfish games (older versions, 1.7, 1.8, 1.9, 2.0 too, x86 and x64 with 1 Core, or an URL ... I will search a little bit start of next week.

Have a nice weekend!

Best
Frank