Hi,
some experiments to share:
- Easy levels (expert mode) and Elo levels (Comfort mode) are indeed the same ones, and changing the setting in one mode changes as well the value in the other mode (ok - not that useful an information
)
- I tried to find some relationship between Comfort levels (adaptive ones) and some Expert mode levels:
1) simulating very weak play using the Ketterling initiative test, the Play & Win level draws quickly (that is kind of it!) while Fun0 and Easy0 play very different moves. Then using adaptive level Comfort Normal, the program achieved mate after playing 33 moves. Level Easy0 provided 22 different moves out of 33, so the Normal adaptive level did not drop to something close to Easy0 (Elo 1000) It plays much differently.
I then tested changing the style (expert mode), I used User defined TRS-aggressive, then back to Comfort mode using Normal level --> the program achieved mate after playing 22 moves (I managed to force first moves in order to have a similar game developped). As game developpment can significantly change the result anyway, I replayed the Ketterling game played by the Normal style/Normal level, forcing the previously achieved moves; and got 14 moves the TRS-aggressive would have played a different way. As a conclusion (to be confirmed after this small test), styles you may select in expert mode seem to apply when using Comfort mode.
2) I simulated very strong play facing Stockfish12 (NNUE enabled), 40/10 20/5 level (average 15s/move), to Comfort mode, Advanced level. My guess was: once stressed, the adaptive program might reach the highest Elo level (8, 2050). The King played around 5 to 7 seconds per move, no ponder. Changing the CPU throttle in Expert mode has no effect. The result is quite interesting, as it played slightly stronger than the maximum Easy 8 / Elo level 8 = 2050 Elo. Here is the game:
[Event "Test KP Comfort levels"]
[Site "Arena"]
[Date "2020.12.05"]
[Round "?"]
[White "Stockfish 12 NNUE"]
[Black "The King Performance lv Comfort Advanced"]
[Result "1-0"]
[ECO "A18"]
[EventDate "2020.12.05"]
[Opening "Slav"]
[Termination "normal"]
[Time "09:39:01"]
[TimeControl "40/600:20/300:20/300"]
[Variation "Exchange, 5.Nc3 e6"]
1.c4 Nf6 2.Nc3 e6 3.e4 d5 4.e5 Nfd7 5.cxd5 Nxe5 6.d4 Ng6 { end of book } 7.dxe6 Bxe6 8.d5 Bf5 9.Qe2+ Qe7 10.Be3 Ne5 { Easy8: Nd7 53215 pos: Ne5 } 11.O-O-O Nbd7 12.h3 h6 13.Nf3 O-O-O { Easy8: a6 86373 pos: O-O-O } 14.d6 cxd6 15.Nd4 Be6 16.f4 Bc4 17.Qe1 Bxf1 18.Nd5 Qe8 { Easy8: Nd3+ 47742 pos: Qe8 } 19.Rxf1 Nc6 20.Nxc6 bxc6 21.Qc3 Kb7 22.Qb4+ Kc8 23.Rfe1 cxd5 24.Bxa7 Be7 25.Rxd5 Kc7 { Qf8 81382: Kc7 } 26.Qa5+ Kc6 27.Qa6+ Kc7 28.Rb5 Rc8 29.Rb7+ Kd8+ 30.Kd2 Qf8 31.Qb5 { Black resigns } 1-0
In the comments, you may read the moves level Easy 8 would have provided, if different ; and the number of analyzed positions a normal level needed, to change from the Easy 8 move to the one played by the adaptive Advanced level move. And obviously, each of these discrepancies were "solved" with more analyzed positions than the Easy 8 level may reach = 32000 (apart from endgame phase, which should not apply in this short game). Average number is close to 67000. As a conclusion, the Advanced adaptative level is able to reach as high as more than 2050 Elo.
Kind regards,
Eric