AlphaZero Chess is not that strong ...

[mjlef]

Because hash size is going to have a large effect using that search scheme.

It seems to be time to fixed depth searches. But with LazySMP, time to depth suffers while the goes up.

It would be really nice to get to the truth on this. HG Muller claimed that going from 1GB to 32 GB only adds 7.5 elo which is quite a bit below what I had guessed (25 elo).

A friend is running something as close as we can get with available hardware. He should post results here once it is done. It will take several days.

[Ozymandias]

I was looking for books from Sedat Canbaz but all his files disappeared : http://www.talkchess.com/forum/viewtopic.php?t=54634

It moved.

I tried the polyglot version of Cerebellum, unfortunately, in some openings it just goes for three-fold repetition.

Polyglot doesn't keep a record of the moves played, it just stores positions. That's one of the reasons why this issue was opened.

[hgm]

Note that when a book contains loops in positions where a draw is not desired, it should be considered an error in the book data, and not an error of the format or probing software. It makes no sense to first repeat a position, and only then play an alternative move because you condider it better than a draw. It should have played that alternative move in the first place.

So the proper solution is to cure this during book creation, and not in the probig code.

[Ovyron]

Note that when a book contains loops in positions where a draw is not desired, it should be considered an error in the book data, and not an error of the format or probing software. It makes no sense to first repeat a position, and only then play an alternative move because you condider it better than a draw. It should have played that alternative move in the first place.

So the proper solution is to cure this during book creation, and not in the probig code.

But it has its advantages. The reason Brainfish + Cerebellum goes to 2-fold repetition is because, if the opposing engine doesn't go for the repetition, it's going to get an even better score than if it played the alternative to the repetition in the first place.

I used to think like you, but Alpha Zero was also going for 2-fold repetition before playing the alternative, and who knows how many of its moves expected Stockfish to try to repeat a position, but Stockfish didn't so A0 was better playing for it than playing the alternative on the first place, but we couldn't tell because Stockfish wrongly thought the repetition was 0.00, and wrongly thought it had the edge.