Werewolf wrote: ↑Wed Mar 02, 2022 10:32 am
Will we see multi-threaded Rebel one day?
Maybe you did not mean that Werewolf but there is of course already Toga IV (1.0) See also Ed's forum: https://prodeo.actieforum.com/t638p125-my-first-nnue (page six)
But Toga really has lots of pruning methods and other modifications that quite possibly you don't even need when you have Rebel's NNUE evaluation. Only the Multi variation and multithreading/MP are really interesting, Ed has said. I would add the aspiration search as well but it is possible Fabien's original infinite window search is better for Rebel or CSTal for that matter. It is not known. Well maybe Ed has tested it. Also there are already some Rebel ideas in Toga IV. I find Fruit much more elegant than Toga. The small parts I understand of it. Like Senpai later for a multithreaded engine. And Fabien's programming was always meant more fore a positional playing style. That was lost a bit with all the search modifications in Toga, by I don't know how many amateur programmers in the past (some were actually truckdriver by profession, or there was Toga Mara that had all sorts of strange geometrical relations in the piece values. Like Voodoo it seemed)
Fruit is more suited for Rebel.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
I found, after lots of tries and errors (but not systematized testing), that switching off hash pruning gives a much more reliable analysis in the case of Toga IV. In general, with the previous Rebel net, an increase in piece activity and King Safety should also be beneficial. The engine is clearly not tuned for the use with NNs, and there is probably 30-50 Elos to be gained with adequate settings.
Werewolf wrote: ↑Wed Mar 02, 2022 10:32 am
Will we see multi-threaded Rebel one day?
Maybe you did not mean that Werewolf but there is of course already Toga IV (1.0) See also Ed's forum: https://prodeo.actieforum.com/t638p125-my-first-nnue (page six)
But Toga really has lots of pruning methods and other modifications that quite possibly you don't even need when you have Rebel's NNUE evaluation. Only the Multi variation and multithreading/MP are really interesting, Ed has said. I would add the aspiration search as well but it is possible Fabien's original infinite window search is better for Rebel or CSTal for that matter. It is not known. Well maybe Ed has tested it. Also there are already some Rebel ideas in Toga IV. I find Fruit much more elegant than Toga. The small parts I understand of it. Like Senpai later for a multithreaded engine. And Fabien's programming was always meant more fore a positional playing style. That was lost a bit with all the search modifications in Toga, by I don't know how many amateur programmers in the past (some were actually truckdriver by profession, or there was Toga Mara that had all sorts of strange geometrical relations in the piece values. Like Voodoo it seemed)
(some were actually truckdriver by profession, or there was Toga Mara that had all sorts of strange geometrical relations in the piece values. Like Voodoo it seemed) 