Never thought that Alliestein was in the same level with LCZero and Stockfish !

[Ovyron]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

[Chessqueen]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

[Chessqueen]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

[mwyoung]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

I don't see much difference in how Lc0.exe and Allie.exe run the NN. Other then Allie does not know how to count nodes correctly compared to Lc0. As the search depths are the same regardless of how Allie is displaying NPS. If one gets into the super final. It is more random chance then one is better then the other.

[Chessqueen]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

I don't see much difference in how Lc0.exe and Allie.exe run the NN. Other then Allie does not know how to count nodes correctly compared to Lc0. As the search depths are the same regardless of how Allie is displaying NPS. If one gets into the super final. It is more random chance then one is better then the other.

I believe that Alliestein will face Stockfish unless Stockfish Crash one more time in the last two games

[mwyoung]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

I don't see much difference in how Lc0.exe and Allie.exe run the NN. Other then Allie does not know how to count nodes correctly compared to Lc0. As the search depths are the same regardless of how Allie is displaying NPS. If one gets into the super final. It is more random chance then one is better then the other.

I believe that Alliestein will face Stockfish unless Stockfish Crash one more time in the last two games

I agree, but it makes no difference. Allie and Lc0 are pretty much the same. So the match up will be NN vs A/B again. It makes little difference in the name of the .exe file Allie.exe or Lc0.exe. I give it even money on who is better this year. With my testing showing SF and Lc0 or Allie to close to call with the latest SF and the best networks. With a Leela Ratio of 1.1 in testing.

It will be funny if SF does crash. And we have a match up of Lc0, and a Lc0 clone.

[Jouni]

SF wins league even with 2 crashes! What happened to NN revolution? BTW both SF crashes did happen when there were 20 or 40 moves with same evalution. Maybe this gives a hint about the reason?

[Chessqueen]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

I don't see much difference in how Lc0.exe and Allie.exe run the NN. Other then Allie does not know how to count nodes correctly compared to Lc0. As the search depths are the same regardless of how Allie is displaying NPS. If one gets into the super final. It is more random chance then one is better then the other.

According to Gonzochess75 at the moment Alliestein only use ab, whereas Leela uses ab+ MTCS + Minimax which makes them totally different, but in two days for the superfinal he will add MTCS + Minimax , which will makes Alleinstein similar to Leela in some ways, it might help and it might NOT, since some MTCS implementation sometimes slow it down and it might even makes Alliestein worse. Hope he does several testings against the current Alliuestein before he decides to use the latest Alliestein version

[Chessqueen]

I don't understand, that clever people at fishcooking are fighting to get 1% more speed!

Maybe they're desperate to beat the Networks?

Stockfish made a great comeback, but there are only 3 more games for each engine to determine the two best engines and Alliestein is having a great tournament regardless of stockfish bug https://tcec-chess.com/

WOW what a comeback by Stockfish, it is tie with Alliestein with two more games to go before playing foe the Superfinal against Alliestein Version 5.0 with a newer Network https://tcec-chess.com/

I don't see much difference in how Lc0.exe and Allie.exe run the NN. Other then Allie does not know how to count nodes correctly compared to Lc0. As the search depths are the same regardless of how Allie is displaying NPS. If one gets into the super final. It is more random chance then one is better then the other.

According to Gonzochess75 at the moment Alleistein only use ab, whereas Leela uses ab+ MTCS + Minimax which makes them totally different, but in two days for the superfinal he will add MTCS + Minimax , which will makes Alleinstein similar to Leela in some ways, it might help and it might NOT, since some MTCS implementation sometimes slow it down and it might even makes Alliestein worse. Hope he does several testings against the current Alliuestein before he decides to use the latest Alliestein version

Allie's search is a combination of MCTS and Minimax. The MCTS is very much like you would find in AlphaZero or LC0 as it was taken from the same papers. In addition to MCTS, Allie also does a minimax based backup of the tree whenever a new MCTS batch is evaluated by the NN. This minimax backup is then used to modify the Q values of the nodes in the tree that were either 'min' or 'max'. This modification is a further averaging of the MCTS based backups. So Allie uses MCTS for playouts and for back prop in much the same way as A0 and Lc0
Allie also does a minimax-based back prop to supplement that mcts based back prop
Gomzo had tried many, many, different ways of adding newer AB/Minimax to the algorithm, but so far nothing that has successfully gained elo over the above. Finally, he noted that in all of jos testing of Allie (pure MCTS) vs Allie (MCTS+Minimax) the latter has scored significantly higher so he is confident that it does improve elo at least in self play.

[corres]

In general the chess power of an NN type engine is determined by the used NET. Particularly if the policy head - what helps engine during search - is in the given NET and this is case of Leela and her derivatives.
So if Allie as a program made for utilizing of Leela`s NET works well her chess power will be the same as Leela has.
Because a lot of important parts of Leela is in Allie (see statements of crem) for a good programmer is not a heavy task to write a good program what more or less differs from Leela and gives the same chess power.