Engine's "ELO" revisited

[Ponti]

My computer's cpu is a Ryzen 1800X. It is not "state of the art" but it is a beast when playing chess at full strengh.

It is hard to choose a "weak" engine based on current rating lists because the hardware used are not the same. So if I pick an "1900" engine, with my cpu it plays stronger, certainly (most of the lists available use older hardware).

Those rating lists are "engine-engine" lists and that's not a "human ELO". Stockfish 10 at full strengh playing in a human tournament with GMs only, would end #1 with an almost perfect score and an incredible rating-performance. Maybe it would be 4000 ELO (or more) in a "human" list.

Is it possible to make such an "ELO" list for engines based on engine-human games ? (I mean strong players, Fide master (FM) level, at least)

[Rebel]

If your goal is to find a comp opponent at your level you might give ProDeo 2.4 a try.

It starts playing at an estimated elo rating of 1500 and you can win or lose elo points.

http://rebel13.nl/prodeo/prodeo24.html

[Ponti]

If your goal is to find a comp opponent at your level you might give ProDeo 2.4 a try.

It starts playing at an estimated elo rating of 1500 and you can win or lose elo points.

http://rebel13.nl/prodeo/prodeo24.html

Mr. Ed,

I can assure you, I've played many games against Prodeo 2.4. At blitz it is by far better than me (I'm somewhere in the 1800-1900 range) and it gives me a much lower rating. I did not try it at slow tournament time controls, though. I did lost many games on time... but it is difficult to win if Prodeo gets a minimal material advantage.

[Rebel]

Thanks for feedback, you are the first one

I think you are right, I should adapt the engine elo rating also based on time control, I remember the difficulties GM's had (at the time of REBEL) too well and playing these guys at tournament control was a whole different story.

Will think about an update.

If you are speaking about "blitz", what time controls did you play?

[Ponti]

Most of the time I play 5 min+ 3 sec./move time increment.
Last 8 games I lost 7-1 to Prodeo 2.4.

I also want to try Bob's Crafty 25.3 , but I think it is hard to set the crafty.rc file to a similar level.

Shredder's adaptive function is good, I'm playing hard games with an old Android tablet (Motorola Xoom, 1st edition).

Things like tablebases and opening books also matters here. It's not fair to reach an endgame against an engine with tablebases, so I almost all the time turn them off.

I don't know why, but some engines without opening book choose 1...Nc6 against 1.e4, so maybe that's the best reply to it...

[jdart]

Arasan has variable strength (settable in its native GUI from 0-100). I am pretty sure you could find a setting that gave you a fairly even match.

--Jon

[Ovyron]

I'm going to make a commercial plug in here for Chessmaster 9000. It has an astounding number of chess personalities and covers all levels of play, up from the random mover to its maximum strength. I would be shocked if you couldn't find a personality to play that suits you, and it entices you to play stronger ones as you master the old ones which you'd do as you practice.

[Ponti]

I really enjoy CM - using Grandmaster Edition (version = XI ?), one of the best !

[Ponti]

I don't know if Josh at 9 years old was that strong in endgames - CM personality is almost a fide master (but plays badly in the opening).

I apreciate that "personalities" approach. Used to play fast games against Hiarcs (CB version 9) and Rebel DOS and it was like a dream. Well, Hiarcs beats me badly. At least against Rebel I had a chance to win or draw some games.

[MikeB]

My computer's cpu is a Ryzen 1800X. It is not "state of the art" but it is a beast when playing chess at full strengh.

It is hard to choose a "weak" engine based on current rating lists because the hardware used are not the same. So if I pick an "1900" engine, with my cpu it plays stronger, certainly (most of the lists available use older hardware).

Those rating lists are "engine-engine" lists and that's not a "human ELO". Stockfish 10 at full strengh playing in a human tournament with GMs only, would end #1 with an almost perfect score and an incredible rating-performance. Maybe it would be 4000 ELO (or more) in a "human" list.

Is it possible to make such an "ELO" list for engines based on engine-human games ? (I mean strong players, Fide master (FM) level, at least)

The Honey family of engines ( they are all SF derived, and each a play Elo function that were designed to match FIDE Elo levels ( roughly - nothing could be exact) and weakens engines by reducing NPS and by playing witty a variety. All 10 engines are just a click away. One engine in particular might suit your needs - Weakfish , Stockfish derived with none of the "search" smarts - caps out about 2400/2500 Elo and also has the aforementioned tools to play weaker. Use with the UCI _Sleep function checked for best results and to avoid instantaneous moves, it will auto sleep appropriately for the for time control selected. It also has function to adjust your normal SF nps to your processor to account for processor speed differences.

https://github.com/MichaelB7/Stockfish/releases/tag/XR7