I wonder what is the CCRL rating of the leela odd bots against engines

[Uri Blass]

I think stockfish is not the best engine to analyze odd games because it does not care about playing the best winning move because it does not add elo.

If I analyze games against Leela Odds bot with engines to see where I went wrong I want engines that prefer simplifications and I think for example Wasp4.50 with no NN is better than Stockfish.
Of course I need to analyze with an engine that usually beat the Leela bot but best engine is not important.

Here is for example a position from one of my losses against LeelaRookOdds

[d]1nb2rk1/ppq1ppb1/5np1/7p/3QB2P/2B3P1/PPPN1P2/R3K1NR w KQ - 6 12

I played 12.Qe5 with the idea to trade queens but after Qb6 understood that it is wrong to capture b8
I believe the best move was 12.Qxa7 when Stockfish believes that the best move is 12.Ngf3

Maybe stockfish after many hours can play 12.Qxa7 and maybe not(I decided to check it with 4 options and scores of 12.Nf3 12.Qe5 and 12.Qxa7 are close. I remember that analysis by Wasp suggested 12.Qxa7 is better by a big margin and I know this engine like simplifications so it is probably not only about the pawn at a7.

I guess the odds bot including the knight odd bot are going to lose usually against Wasp4.50 in CCRL conditions but I am not 100% sure about it.
Wasp4.50 4 cpu CCRL rating is 3282
https://www.computerchess.org.uk/ccrl/4 ... 4-bit_4CPU

[lkaufman]

I think stockfish is not the best engine to analyze odd games because it does not care about playing the best winning move because it does not add elo.

If I analyze games against Leela Odds bot with engines to see where I went wrong I want engines that prefer simplifications and I think for example Wasp4.50 with no NN is better than Stockfish.
Of course I need to analyze with an engine that usually beat the Leela bot but best engine is not important.

Here is for example a position from one of my losses against LeelaRookOdds

[d]1nb2rk1/ppq1ppb1/5np1/7p/3QB2P/2B3P1/PPPN1P2/R3K1NR w KQ - 6 12

I played 12.Qe5 with the idea to trade queens but after Qb6 understood that it is wrong to capture b8
I believe the best move was 12.Qxa7 when Stockfish believes that the best move is 12.Ngf3

Maybe stockfish after many hours can play 12.Qxa7 and maybe not(I decided to check it with 4 options and scores of 12.Nf3 12.Qe5 and 12.Qxa7 are close. I remember that analysis by Wasp suggested 12.Qxa7 is better by a big margin and I know this engine like simplifications so it is probably not only about the pawn at a7.

I guess the odds bot including the knight odd bot are going to lose usually against Wasp4.50 in CCRL conditions but I am not 100% sure about it.
Wasp4.50 4 cpu CCRL rating is 3282
https://www.computerchess.org.uk/ccrl/4 ... 4-bit_4CPU

The Leela Odds Bots don't do as well giving odds to standard (no NN) engines as they do against humans of equal strength in standard chess, but they do better than engines like Stockfish or Komodo or normal Leela do. Wasp 4.5 is clearly too strong for even knight odds in Rapid play, but perhaps on one thread in bullet chess or with limited nodes it might be a suitable test opponent for the bots. I agree that it should be better to use for analyzing odds games than an engine that doesn't try hard to simplify when way ahead.

[Uri Blass]

I think stockfish is not the best engine to analyze odd games because it does not care about playing the best winning move because it does not add elo.

If I analyze games against Leela Odds bot with engines to see where I went wrong I want engines that prefer simplifications and I think for example Wasp4.50 with no NN is better than Stockfish.
Of course I need to analyze with an engine that usually beat the Leela bot but best engine is not important.

Here is for example a position from one of my losses against LeelaRookOdds

[d]1nb2rk1/ppq1ppb1/5np1/7p/3QB2P/2B3P1/PPPN1P2/R3K1NR w KQ - 6 12

I played 12.Qe5 with the idea to trade queens but after Qb6 understood that it is wrong to capture b8
I believe the best move was 12.Qxa7 when Stockfish believes that the best move is 12.Ngf3

Maybe stockfish after many hours can play 12.Qxa7 and maybe not(I decided to check it with 4 options and scores of 12.Nf3 12.Qe5 and 12.Qxa7 are close. I remember that analysis by Wasp suggested 12.Qxa7 is better by a big margin and I know this engine like simplifications so it is probably not only about the pawn at a7.

I guess the odds bot including the knight odd bot are going to lose usually against Wasp4.50 in CCRL conditions but I am not 100% sure about it.
Wasp4.50 4 cpu CCRL rating is 3282
https://www.computerchess.org.uk/ccrl/4 ... 4-bit_4CPU

The Leela Odds Bots don't do as well giving odds to standard (no NN) engines as they do against humans of equal strength in standard chess, but they do better than engines like Stockfish or Komodo or normal Leela do. Wasp 4.5 is clearly too strong for even knight odds in Rapid play, but perhaps on one thread in bullet chess or with limited nodes it might be a suitable test opponent for the bots. I agree that it should be better to use for analyzing odds games than an engine that doesn't try hard to simplify when way ahead.

I think many engines do not know to simplify the position and trade pieces so I am surprised if Leela Odds bot can do better against engines relative to humans of equal strength(assuming you do not mean specific engines that know to simplify).