Whiskers wrote: ↑Thu Jan 02, 2025 4:11 am
I’m 90% sure that Patricia had a bug where some GUIs would explicitly set both skill level AND uci_elo instead of leaving the other one at default, which means you might have been playing full strength Patricia!
Thx, I was aware of that (here, UI overwrites skill setting) and unfortunately I am that weak. Ie. I play with skill level 1 and in Acid Ape Chess set the engine strength to zero. Tested it in Patricia vs. SF matches and at that settings it reliably loses against sf 20 to 30 moves in.
tapio wrote: ↑Thu Jan 02, 2025 9:14 pmNice options. Is uci_elo related to eval-noise? Are they independent? Do they stack up? TIA!
You can set eval-noise and speed throttle manually (the CPU Speed Throttle checkbox must be active for the latter), or do it summarily via UCI_Elo. UCI_Elo will override a manual setting if you try to set both.
Ras wrote: ↑Thu Jan 02, 2025 10:17 pmYou can set eval-noise and speed throttle manually (the CPU Speed Throttle checkbox must be active for the latter), or do it summarily via UCI_Elo. UCI_Elo will override a manual setting if you try to set both.
a) "both" - uci_elo + eval_noise? Or eval_noise+speed throttle?
b) But isn't uci_elo always set - to some value? Same as eval-noise? What does "set both" mean?
Aww, it's always confusing dealing with internal settings and the external uci_elo set by the gui.
tapio wrote: ↑Fri Jan 03, 2025 7:04 ama) "both" - uci_elo + eval_noise? Or eval_noise+speed throttle?
eval_noise/speed + uci_elo. What uci_elo does under the hood is controlling eval_noise and speed throttle.
b) But isn't uci_elo always set - to some value?
No, only if uci_limitstrength is also set. The GUI will manage this if you use the Elo slider. E.g. the Shredder GUI has a "Limit Strength" checkbox besides the Elo slider.
Same as eval-noise?
That is set to 0 by default.
What does "set both" mean?
Using both the Elo slider that causes uci_elo and uci_limitstrength to be set, as well as configuring eval_noise/speed manually.
No, only if uci_limitstrength is also set. The GUI will manage this if you use the Elo slider.
Great, thank you, that explains it. uci_limitstrength probably is set to off by GUIs if slider/whatever element is at full strength. Acid Ape Chess has a slider per engine, with values mapped to 0-100%.
You might be looking at engines below 2800/2400/2000 by selecting them from the CCRL list.
Komodo does a good job at lowering it's level using skill parameter. Stockfish lower levels is generating very imbalanced play.
To have fun, I recommend not playing >2500 engines, except if you are GM.
tapio wrote: ↑Tue Dec 31, 2024 9:28 am
Any experiences with more humanly playing engines? I'm a weak player capable to even lose against Rodents weak School personalities. Patricia 4 at lowest skill still beats me. It's a concentration issue, my blunders can be horrible.
I think Rodent has some code to deal wirh human blunders. It usually instantly counter blunders, which I always see after a proper engine analysis. I don't like the instant counter blunder.
Do you know other engines with have some interesting blunder/bad movement management included? I wish developers put more thought into such things instead of boring elo achievements withgames which nohumanbeing can judge anyway.
Check out my engine. I'm not actively developing it for strength improvement, but I keep tinkering with its limit-strength feature and sparring against it. See The MadChess UCI_LimitStrength Algorithm for a detailed explanation of how I weaken the engine. I don't know how well calibrated the Elo is. Though you can adjust formulas via a JSON file if you want fine-grained control.
I'm a novice player, about 1000 Elo at rapid and 700 Elo at blitz on chess.com. At those ratings, MadChess gives me a fun game. It doesn't play like a super GM then hangs a piece. It plays at a fairly consistent level throughout the game.
had a quick look (again) at Mad chess indeed plays sloppy chess (*) at weakened
level, which is good ! (to simulate human amateur chess).
One small comment: without opening book it imo plays a bit (too)
low in the opening (tested at rapid level); and when using a book,
eg a human amateur style book, it imo shouldn't play immediately
(like almost all engines do) but after a few seconds, say up to 30 secs
or so. With time gradually increasing as it becomes more difficult
for the 'human' to digup info from memory (and/or checking
the move anyway to prevent huge opening blunders).
(*) it even used a (very rare) gambit with White in the French
1.e4 e6 d4 d5 3 Bf4?! (played a few times below 2000, but does not
have a gambit name; so i would suggest French 'Mad gambit'
(after dxe4 4.Nc3 is best, but in practice 4.f3!? has scored better)
jefk wrote: ↑Sun Jan 05, 2025 1:25 pm(like almost all engines do) but after a few seconds, say up to 30 secs
You should definitely look at LucasChessR - it's IMO the best UI we have today, and actively developed. It can influence time from UI side. The option is IIRC called "normalize time". The UI is also capable to influence playing behaviour of the engine, eg not playing the best move.
Ie. I play with skill level 1 and in Acid Ape Chess set the engine strength to zero. Tested it in Patricia vs. SF matches and at that settings it reliably loses against sf 20 to 30 moves in.
