TalkChess.com

with a book and tablebases forget the word "zero" its not that a pure self learning,looks like a "cheat" and means the project "zero" failed.

crem wrote: ↑Sat Jul 21, 2018 5:43 pm

jp wrote: ↑Fri Jul 20, 2018 6:47 am

yanquis1972 wrote: ↑Thu Jul 19, 2018 11:50 pm if there was no book, it seems to me leela did awfully well in the opening.

What do you think of as "awfully well"? The Leela guys seem to be making the opposite claim. e.g. "Leela was handicapped by being -3 out of every opening" (not a real quote).

Who said that? That's not really true, Leela does well in opening, and in all WCCC games she went out of openings with both good self-eval and eval from other engines.

Out of entire Lc0 team, I think that an author of the blog is the only person who think that Lc0 would benefit from opening book, so there may be hints to that in the blog.

Yeah, Dietrich Kappe is just one opinion & not "the Leela guys". But it was way more than "hints".

"Leela is playing without a book, so she’s vulnerable to better booked opponents."
"Again, no opening book, and it had its consequences."
"Imagine what Leela could have done with an equal or better position. Right now she is no match for theory."
"To recap: Leela’s lack of opening book combined with some fairly strong engines running on very beefy hardware led to some difficult games."

stavros wrote: ↑Sun Jul 22, 2018 1:07 am with a book and tablebases forget the word "zero" its not that a pure self learning,looks like a "cheat" and means the project "zero" failed.

Who cares the name zero. This is chess engines competition. Any eligible benefits should be implemented for best performance(e.g Tablebase access)

Nay Lin Tun wrote: ↑Sun Jul 22, 2018 1:29 am

stavros wrote: ↑Sun Jul 22, 2018 1:07 am with a book and tablebases forget the word "zero" its not that a pure self learning,looks like a "cheat" and means the project "zero" failed.

Who cares the name zero. This is chess engines competition. Any eligible benefits should be implemented for best performance(e.g Tablebase access)

I would even go further. Ignoring human expert knowledge is ok because it can be wrong or at least imprecise, and indeed it would be somewhat incompatible to the project idea. But ignoring perfect knowledge from endgame tablebases would just be silly. I see EGTBs almost on the same level as the chess rules. And ignoring the option to save time by using an opening book would be silly as well, provided the opening book is mainly based on own analysis and therefore not on human expert knowledge. Why letting the program always calculate the same first move, the same reaction on the first enemy reply etc.?

Using an opening book based on current opening theory can be seen controversial: on the one hand it is obviously incompatible to the project idea but on the other hand such a book may be need for one practical reason: to ensure a variety in the opening moves in order to avoid many repeated games, especially in a match vs. one single opponent. You can't derive sufficient information about playing strength from only few different games, after removing all duplicates. Whether a book for variety purposes must be based on human opening theory or not is another issue.

stavros wrote: ↑Sun Jul 22, 2018 1:07 am with a book and tablebases forget the word "zero" its not that a pure self learning,looks like a "cheat" and means the project "zero" failed.

I do not see the problem with book and tablebases.

1)A book does not have to be based on opinion of human players and it can be based on analysis by LCzero before the games in order to save time on the clock.
2)Tablebases are derived from zero and not from human games.

well.. A0 made it without opening book and without tablebases

but that sort of knowledge can only make an engine stronger. An opening book is like 10 million training games and having tablebases definitely saves neuron network access time

Uri Blass wrote: ↑Sun Jul 22, 2018 10:25 pm

stavros wrote: ↑Sun Jul 22, 2018 1:07 am with a book and tablebases forget the word "zero" its not that a pure self learning,looks like a "cheat" and means the project "zero" failed.

I do not see the problem with book and tablebases.

1)A book does not have to be based on opinion of human players and it can be based on analysis by LCzero before the games in order to save time on the clock.
2)Tablebases are derived from zero and not from human games.

Provided 1) is under those conditions it would be OK, but not otherwise.