Rating List Links

[karakaniec]

JCR Ranking 11.09.2012 (64 bit) Temp 3' +1" and 5'
1 Critter 1.6a 64-bit 3022 116 116 22 66% 327 41%
2 Houdini 1.5a x64 2956 44 44 146 59% 303 48%

More: http://chessengines.blogspot.com/2012/0 ... 4-bit.html

[karakaniec]

Rank Name Elo + - games score oppo. draws
1 Houdini 2.0c Standard x32 3207 126 126 20 70% 583 30%
2 Critter 1.6a 3128 25 25 456 55% 612 48%
3 Komodo 5 x32 3123 86 86 38 57% 589 45%
4 Strelka 5.5 3111 26 26 428 54% 583 49%
5 IvanHoe 47c+ GH w32 3086 119 119 20 45% 607 40%
6 Tankist 3.1 x32 3071 33 33 270 50% 566 49%
7 Tactico 3350b ProLG x32 3066 17 17 1032 58% 507 46%
8 Stockfish 2.2.1 3062 54 54 96 43% 608 52%
9 Fire 1.5xTreme_n0_SSE2_NS 3054 72 72 48 53% 538 69%
10 Sting SF 1.0 3051 71 71 77 62% 407 30%

Top 120 engines (games, book, etc.)

[karakaniec]

Jurek Chess Engines Ratind 03.2013:
https://sites.google.com/site/chessengi ... er01032013

[karakaniec]

Jurek Chess Engines Rating Live 18.03.2013

[karakaniec]

JCER Live:
http://chessengines.blogspot.com/2013/0 ... 32013.html

[pedrox]

I can add a philosophical question that has recently entered been mentioned with the R4 release (and testing therein).

Why test "ponder off" while simultaneously allowing the engine to manage time? Why not just test old-style lightning chess with repeated "go movetime 10000" commands or the like? I can agree that "ponder on" tests the whole engine (at least if you don't attach the book to the engine) at chess-playing, but why bother to test time management once you've made the leap to turning ponder off? I think one author said that "ponder off" was rather arbitrary, and rather like saying "no qsearch". Given that time management is variously claimed to be worth as much as 20 ELO (or more), it seems that the "ponder off" lists might want to exclude this facet, especially if the idea of "ponder off" testing is to give an idea of how good of analysis to expect from the engine. Then again, some engines seem to have a different schema in analysis versus gameplay, so maybe no metric is perfect.

I think more or less the same. For example it is unfair to play on one computer with 4 processors a engine with 1 CPU and another with 4 CPU and ponder is off.

One reason for playing ponder off is have one computer with one processor, even with 2 processors may be justified ponder off, with 3 processors you can leave one for each engine and another for the GUI and the rest.

[Kajordzak]

CCRL 40/4: http://computerchess.org.uk/ccrl/404/
CCRL 40/40: http://computerchess.org.uk/ccrl/4040/
IPON: http://www.inwoba.de/
SWCR: http://www.amateurschach.de/
CEGT: http://www.husvankempen.de/nunn/

Does anybody have a proof that rating list ELO correlates to chess playing skill?

Or that the relationship between rating list ELO and chess playing skill is in any way linear?

Or that changes in rating list ELO map negatively or positively to decreases/increases in chess playing skill?

Or that rating lists composed from multi-games between similar machine entities actually measure anything useful at all?

Why not just throw these lists into the poubelle?

Rankings are not created from the ceiling, I trust that arise from games played. There are so necessary and meaningful ...

First tests and ranking SSDF began with the outcome of the games of chess programs against the people, then this is not repeated for the SSDF, and of course it is a great pity. The exception was a cycle of AEGON, but now chess players are reluctant to play against chess programs.

Your (respectfully) is similar to the reluctance of the former action against the SSDF, when the threats against the SSDF had a long time to encode the name of the programs in the test results and ranking!

[karakaniec]

Jurek Chess Engines Rating Live - 08.06.2013

[karakaniec]

Jurek Chess Engines Rating Live - 18.08.2013

[karakaniec]

Jurek Chess Engines Rating live – 13.10.2013 (BayesELO) Temp: 3' + 3"