This is the result of Sfetan's EAS tool based on 8000 training games played at bullet time control.
As an addition I wrote a tool that only looks at sacrifices engines make and (of course) a queen sacrifice is a lot more worth than a pawn sacrifice. We get -
This is the result of Sfetan's EAS tool based on 8000 training games played at bullet time control.
As an addition I wrote a tool that only looks at sacrifices engines make and (of course) a queen sacrifice is a lot more worth than a pawn sacrifice. We get -
We can see a couple of remarkable changes, SF miles ahead, Slowchess was 4th and now almost last.
Both sacrifices and moves to win, all else equal, are proportional to Elo. To get at the "all else equal" components, you would need to normalise for Elo.
EAS, btw, is highly volatile and needs possibly 20,000 games before it gets stable. In general, I think we are "interested" in sacrifices the engine chooses on evaluation grounds, and not because it sees a win at the end of the line (that's not really a sacrifice, imo). The current tool for edfining a sac doesn't differentiate between eval sacs and search sacs. Obviously search sac count will increase with Elo, which is why SFs are scoring high there.
One differentiation method could be to measure the number of moves the sac condition held for, the greater the number, the more likely it's an evaluation sac. You need a PGN parser with more functionality than the one used at the moment.
Nezhman and Ghppn like this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Sun Jun 18, 2023 5:12 pm
This is the result of Sfetan's EAS tool based on 8000 training games played at bullet time control.
As an addition I wrote a tool that only looks at sacrifices engines make and (of course) a queen sacrifice is a lot more worth than a pawn sacrifice. We get -
We can see a couple of remarkable changes, SF miles ahead, Slowchess was 4th and now almost last.
Both sacrifices and moves to win, all else equal, are proportional to Elo. To get at the "all else equal" components, you would need to normalise for Elo.
EAS, btw, is highly volatile and needs possibly 20,000 games before it gets stable. In general, I think we are "interested" in sacrifices the engine chooses on evaluation grounds, and not because it sees a win at the end of the line (that's not really a sacrifice, imo). The current tool for edfining a sac doesn't differentiate between eval sacs and search sacs. Obviously search sac count will increase with Elo, which is why SFs are scoring high there.
One differentiation method could be to measure the number of moves the sac condition held for, the greater the number, the more likely it's an evaluation sac. You need a PGN parser with more functionality than the one used at the moment.
Correct observation and at bullet level SF performs at its best.
The most simple solution to measure sacs is to use elo pools with engines of about equal elo.
Ghppn likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Sun Jun 18, 2023 5:23 pm
Example with 250K games of the gambit rating list.
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 6:51 am
All-time Top10 EAS-Ratinglist
This ratinglist contains the 10 engines / engine-versions (later and stronger versions of engines often play less aggressive!) with the highest EAS-score out of my full ratinglist (all tested engines since 2020, except Stockfish Dev-versions). If the same engine would be part of this Top10 list with different versions, the engine-version with the highest EAS-Score is chosen. The full EAS-Ratinglist can be found below the SPCC full-ratinglist - just click here and scroll down...The Elo-numbers were taken from the SPCC full-ratinglist at the time, this Top10 list was updated...
Latest update: 2023/06/12: Fire 9 (strongest engine (3508 SPCC-Elo) in this All-time Top10 EAS-Ratinglist)
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 8:23 am
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool. Velvet 4.1.0 gets a very nice EAS score against Rebel 15 but against Rebel 16.2 it gets no score at all because Velvet 4.1.0 never reaches the required minimum of 50 won games.
In theory adding games of 2800-3000 elo pool of engines could produce a total different list, actually it's what the (above) output of the gambit-rating-list is telling you.
Besides making our engines stronger we also are keen on playing style and the EAS tool is useful for that purpose but it only makes sense using top-10 engines.
Ghppn likes this post
pohl4711
Posts : 160 Join date : 2022-03-01 Location : Berlin
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 10:05 am
Admin wrote:
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool.
And that is exactly what I am doing. In my SPCC-Ratinglist, the engines are tested versus other engines in the range of the estimated Elo-strength of the new engine, which is tested. And the games of the SPCC Ratinglist are used by the EAS-Tool for the EAS-Ratinglist. And for the Full EAS-Ratinglist, too (from there, the All-time Top10 EAS-Ratinglist is built). So, the EAS-Ratinglist does not fool anybody.
Ghppn likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 10:27 am
pohl4711 wrote:
Admin wrote:
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool.
And that is exactly what I am doing. In my SPCC-Ratinglist, the engines are tested versus other engines in the range of the estimated Elo-strength of the new engine, which is tested. And the games of the SPCC Ratinglist are used by the EAS-Tool for the EAS-Ratinglist. And for the Full EAS-Ratinglist, too (from there, the All-time Top10 EAS-Ratinglist is built). So, the EAS-Ratinglist does not fool anybody.
Then add the violet text (or something like that) to your website as currently folks might draw conclusions such as that Velvet, Arasan etc. play more aggressive than Stockfish.
Ghppn likes this post
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 11:26 am
pohl4711 wrote:
Admin wrote:
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool.
And that is exactly what I am doing. In my SPCC-Ratinglist, the engines are tested versus other engines in the range of the estimated Elo-strength of the new engine, which is tested. And the games of the SPCC Ratinglist are used by the EAS-Tool for the EAS-Ratinglist. And for the Full EAS-Ratinglist, too (from there, the All-time Top10 EAS-Ratinglist is built). So, the EAS-Ratinglist does not fool anybody.
Understood. I think, though, that what doesn't make sense is to have all engines in the same list, weak and strong, as if the EAS values compare.
Sargon comes in at 16th place out of 1011 engines. But Sargon is not "aggressive, it's not anything, it's just a very primitive fixed look-ahead bean counter, 1980 technology, simplest of eval functions.
All those high scoring EAS engines with low Elo, simply cannot be compared to high Elo developed engines, on the same scale. I think what would be best (for understanding and ranking) would be to place engines by Elo in separate divisions and then compare within the division.
There also seems to be quite a difference depending on NNUE or not.
Ghppn likes this post
pohl4711
Posts : 160 Join date : 2022-03-01 Location : Berlin
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 11:40 am
Admin wrote:
pohl4711 wrote:
Admin wrote:
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool.
And that is exactly what I am doing. In my SPCC-Ratinglist, the engines are tested versus other engines in the range of the estimated Elo-strength of the new engine, which is tested. And the games of the SPCC Ratinglist are used by the EAS-Tool for the EAS-Ratinglist. And for the Full EAS-Ratinglist, too (from there, the All-time Top10 EAS-Ratinglist is built). So, the EAS-Ratinglist does not fool anybody.
Then add the violet text (or something like that) to your website as currently folks might draw conclusions such as that Velvet, Arasan etc. play more aggressive than Stockfish.
Of course, Velvet 4.1 plays much more aggressive than Stockfish (and Velvet 5.2 still just a little bit).
Ghppn likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 12:22 pm
pohl4711 wrote:
Admin wrote:
pohl4711 wrote:
Admin wrote:
EAS is a great tool, but don't let the list fool you, it only works with engines within a reasonable close elo pool.
And that is exactly what I am doing. In my SPCC-Ratinglist, the engines are tested versus other engines in the range of the estimated Elo-strength of the new engine, which is tested. And the games of the SPCC Ratinglist are used by the EAS-Tool for the EAS-Ratinglist. And for the Full EAS-Ratinglist, too (from there, the All-time Top10 EAS-Ratinglist is built). So, the EAS-Ratinglist does not fool anybody.
Then add the violet text (or something like that) to your website as currently folks might draw conclusions such as that Velvet, Arasan etc. play more aggressive than Stockfish.
Of course, Velvet 4.1 plays much more aggressive than Stockfish (and Velvet 5.2 still just a little bit).
Not not against Stockfish and that's the assumption your current website gives.
Ghppn likes this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Tue Jun 20, 2023 3:28 pm
Here is an idea for your nice tool, split the pgn database into 3-4 elo pools / divisions.
Mclane and Ghppn like this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Addition to Stefan Pohl EAS rating list Sun Jul 02, 2023 11:38 am
Made an EAS list from the 40/15 CCRL games, elo pool 3400+
------------------------------------------------------------------- *** Average length of all won games: 60 moves *** Engine gets bonuspoints, if its avg. won games length is shorter *** Engine gets maluspoints, if its avg. won games length is longer
Ghppn likes this post
Sponsored content
Subject: Re: Addition to Stefan Pohl EAS rating list