But they're still both valuable for analysis.Before we engage with the topic of the day – ‘ What is Stockfish chess?’, it is essential that we first deal with the meaning of the term ‘chess engine’. Honestly, SF is most likely the better of the engines on your system, if you're looking to figure out which one is "best". Leela needs a crazy powerful GPU and to be run with Cuda or similar in order to get anywhere close to the strength of SF. Then there's the difference between CPU and GPU. On the other side, SF is much faster at looking at positions and finding tactics than Leela. SF needs to look further into the future to really figure it out. Technical differences between Leela and SF can sometimes make things look weird.įor example, Leela needs less depth than SF because Leela is better at deciding how good a position is. Both engines will do an equally good job with this. The best way to use them is to figure out why certain moves are bad and to find out what general advantages a certain type of position can give you. The best way to use engines isn't to find the "best" move because we have no idea if there is even such a thing in most cases. But I think as long as you can distinguish between 'small advantage, a lot better, winning and crushing' I'd say comparing numbers is kinda pointless. Your second question (about leela scoring positions with higher absolutz values): I know leela's evaluation representation is different from stockfish' but I can't remember how exactly. In these kinds of positions I think engines aren't really that useful for humans because they usually don't help with coming up with a plan. If they both give different moves but both engines evaluate the position roughly the same way then it's really just a matter of taste. Now when you think about Leela and SF as just extremely strong ultra-grandmasters, then you could say that if they both give the same move, saying it's considerably better than other moves. If you show them a position that has a winning tactic they're always going to suggest the same move. Let's draw a comparison to human GM's: if two GM's are looking at an equal middlegame, then odds are those GM's are often going to suggest different moves even if they have similar ratings. When both engines give moves with similar evaluations I wouldn't really worry about it that much tbf. Leela and SF just gave different ways of evaluating positions and thus have different insights. Is this a bug on my side, or on purpose (for example: this endgame is more favorable materially, but I know I will win faster in this one ?) What's a rule of thumb to compare how deep I should go to get similar level to Stockfish for a given depth ?Īlso, Leela sometimes doesn't put the move with the highest score first. Is that generally true, or how can I interpret that ?Īlso, with Leela, it takes longer to go deep. Thus, I really don't know what to think of my Qf1 move (I think the depth is not relevant, Leela doesn't need to go as deep as she has more intuition)Īlso, I noticed Leela's evaluation is generally smaller in absolute value. In the example below:Stockfish (depth 36): I found that they reasonably often disagree, and don't know what to take away from these analyses.Īre there any tips to use the engine more cleverly ? In which situations should I trust one v. I analyze my games with Stockfish 13 (BMI2: ) and Leela (DNN BLAS: ) on an average windows desktop.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |