Triumviratus 8.1 released

[Rebel]

Hi Francesco,

Below is the result of a similarity test between 23 top engines.

Code: Select all

Engine Stockfish-19 and Triumviratus-8.0 show a similarity of 64.05%
Engine Stockfish-19 and Triumviratus-8.1 show a similarity of 63.74%
Engine Stockfish-18 and Triumviratus-8.0 show a similarity of 63.12%
Engine Stockfish-18 and Triumviratus-8.1 show a similarity of 62.60%
Engine Pawnocchio-1.9.2 and Tarnished-5.0 show a similarity of 61.03%

I can say that 64% with Stockfish-19 is extreme high for a NNUE engine.

Most of the top engines score a similarity in the 40's or low 50's.

AI is great, no doubt.

Hi,

Could you share how the similarity was measured? Which positions, how many, and the time or depth per move?

If it is based on how often the two engines play the same move, I don't think the number means much. Two strong engines agree on the best move in most positions. Pawnocchio 1.9.2 and Tarnished 5.0 also score high in your list. I dont see the conclusions of such results.
Thanks.

Simex is a simple math tool that produces a statistic. It let engines play on their lowest elo level (depth=1) for 8238 positions and stores the moves. Then it counts for every engine the number of equal moves and calculate its percentage. For the vast majority of the 8238 positions counts that are at least 2-3 or more playable moves.

[NathanDrake]

Hi Francesco,

Below is the result of a similarity test between 23 top engines.

Code: Select all

Engine Stockfish-19 and Triumviratus-8.0 show a similarity of 64.05%
Engine Stockfish-19 and Triumviratus-8.1 show a similarity of 63.74%
Engine Stockfish-18 and Triumviratus-8.0 show a similarity of 63.12%
Engine Stockfish-18 and Triumviratus-8.1 show a similarity of 62.60%
Engine Pawnocchio-1.9.2 and Tarnished-5.0 show a similarity of 61.03%

I can say that 64% with Stockfish-19 is extreme high for a NNUE engine.

Most of the top engines score a similarity in the 40's or low 50's.

AI is great, no doubt.

Hi,

Could you share how the similarity was measured? Which positions, how many, and the time or depth per move?

If it is based on how often the two engines play the same move, I don't think the number means much. Two strong engines agree on the best move in most positions. Pawnocchio 1.9.2 and Tarnished 5.0 also score high in your list. I dont see the conclusions of such results.
Thanks.

Simex is a simple math tool that produces a statistic. It let engines play on their lowest elo level (depth=1) for 8238 positions and stores the moves. Then it counts for every engine the number of equal moves and calculate its percentage. For the vast majority of the 8238 positions counts that are at least 2-3 or more playable moves.

Thanks for explaning. I would argue that having similar moves beetwen two strong engines is perfectly normal.

[NathanDrake]

Hi Francesco,

Below is the result of a similarity test between 23 top engines.

Code: Select all

Engine Stockfish-19 and Triumviratus-8.0 show a similarity of 64.05%
Engine Stockfish-19 and Triumviratus-8.1 show a similarity of 63.74%
Engine Stockfish-18 and Triumviratus-8.0 show a similarity of 63.12%
Engine Stockfish-18 and Triumviratus-8.1 show a similarity of 62.60%
Engine Pawnocchio-1.9.2 and Tarnished-5.0 show a similarity of 61.03%

I can say that 64% with Stockfish-19 is extreme high for a NNUE engine.

Most of the top engines score a similarity in the 40's or low 50's.

AI is great, no doubt.

Hi,

Could you share how the similarity was measured? Which positions, how many, and the time or depth per move?

If it is based on how often the two engines play the same move, I don't think the number means much. Two strong engines agree on the best move in most positions. Pawnocchio 1.9.2 and Tarnished 5.0 also score high in your list. I dont see the conclusions of such results.
Thanks.

Simex is a simple math tool that produces a statistic. It let engines play on their lowest elo level (depth=1) for 8238 positions and stores the moves. Then it counts for every engine the number of equal moves and calculate its percentage. For the vast majority of the 8238 positions counts that are at least 2-3 or more playable moves.

Thanks for the explanation. I would argue that having similar moves beetwen two strong engines is perfectly normal.