AndrewGrant wrote: ↑Mon Jan 03, 2022 1:17 pm
I do not test without a book -- I would simply get the same line the majority of the time. I don't think anyone tests without a book, it would make little sense. For analysis reasons, I can't see a high level player not exploring a particular line and doing from the starting position instead.
Unfortunately, you are very wrong here. The opening books are wrong. I understand I lost $ 40.
Uh... okay. Well the engine is trained like any other NNUE, with randomized openings. So it learns via that process opening lines that are actually good?
I think you would be hard pressed to demonstrate a significant difference in opening strength between engines at the same elo as Ethereal.
AndrewGrant wrote: ↑Mon Jan 03, 2022 2:27 pm
Uh... okay. Well the engine is trained like any other NNUE, with randomized openings. So it learns via that process opening lines that are actually good?
I think you would be hard pressed to demonstrate a significant difference in opening strength between engines at the same elo as Ethereal.
BTW, do you plan any update to the commercial version soon, Andrew?
Happy New Year regards
AndrewGrant wrote: ↑Mon Jan 03, 2022 2:27 pm
Uh... okay. Well the engine is trained like any other NNUE, with randomized openings. So it learns via that process opening lines that are actually good?
I think you would be hard pressed to demonstrate a significant difference in opening strength between engines at the same elo as Ethereal.
Unfortunately, you forgot that various openings are weak and contain mistakes. You should do a test of your engine without using an opening book, only then will your engine show the true power of the game. And not by any opening books.
AndrewGrant wrote: ↑Mon Jan 03, 2022 2:27 pm
Uh... okay. Well the engine is trained like any other NNUE, with randomized openings. So it learns via that process opening lines that are actually good?
I think you would be hard pressed to demonstrate a significant difference in opening strength between engines at the same elo as Ethereal.
Unfortunately, you forgot that various openings are weak and contain mistakes. You should do a test of your engine without using an opening book, only then will your engine show the true power of the game. And not by any opening books.
If you have a specific test you have in mind in order to judge the quality of an engine without an opening book, yet still getting different lines of play despite the determinism of the engine, I would like to know more and would be happy to run said test and share the results.
AndrewGrant wrote: ↑Mon Jan 03, 2022 5:48 pm
If you have a specific test you have in mind in order to judge the quality of an engine without an opening book, yet still getting different lines of play despite the determinism of the engine, I would like to know more and would be happy to run said test and share the results.
AndrewGrant wrote: ↑Mon Jan 03, 2022 5:48 pm
If you have a specific test you have in mind in order to judge the quality of an engine without an opening book, yet still getting different lines of play despite the determinism of the engine, I would like to know more and would be happy to run said test and share the results.
I am unable to download any of those PGNs. Can you attest to the amount of variety in the games for each match, given that there is no opening book.
How many unique 4-move openings are there in those 100 game sets each?
Is 4-threads sufficient to create enough variance in otherwise deterministic engines to produce variable play?
Have you done a similar test with opening books, and demonstrated elo differences that cannot be simply attributed to drawishness?
AndrewGrant wrote: ↑Mon Jan 03, 2022 6:41 pm
I am unable to download any of those PGNs. Can you attest to the amount of variety in the games for each match, given that there is no opening book.
How many unique 4-move openings are there in those 100 game sets each?
Is 4-threads sufficient to create enough variance in otherwise deterministic engines to produce variable play?
Have you done a similar test with opening books, and demonstrated elo differences that cannot be simply attributed to drawishness?
I'll try to help in some time. Because I am generating a database of endings for Cerberus draughts - and I don't want to stop it. How can you, please reply to this post Peter is a good man and he helped me.
A huge and cordial request to the author of the program. Please also respect other people who bought your program. And not only selected people with whom you share the program.
