8

I have seen some commentary that implies that Stockfish did not have access to an opening book in its much-reported match with AlphaZero, but I cannot find a conclusive citation, and my inability to do so has me wondering if maybe Stockfish did have an opening book after all.

For example:

GM Hikaru Nakamura...called the match "dishonest" and pointed out that Stockfish's methodology requires it to have an openings book for optimal performance. While he doesn't think the ultimate winner would have changed, Nakamura thought the size of the winning score would be mitigated.

or, from a StackExchange answer:

Stockfish...wasn't allowed to use an opening book which makes it consider many moves in shallow depths.

It's worth noting that even if Stockfish didn't have an opening book for the canonical 100 game match, page 6 of the paper shows that AlphaZero is dominant when you start from a variety of common openings. Nonetheless I wonder why they would handicap Stockfish this way.

  • 1
    Just FYI in April 2019 an updated version of AlphaZero was run against "a variant of Stockfish that uses a strong opening book". The results were more favourable to stockfish, but AlphaZero was still the overall winner by a convincing margin. – fostandy Feb 16 at 13:06
9

Both Nakamura and the StackExchange answer were correct - Stockfish didn't use an opening book. I think Google did that to make a fair comparison as AlphaZero wasn't using an opening book as well.

EDIT

I didn't see anything about opening book for Stockfish in the paper. I'm not the only one.

http://talkchess.com/forum/viewtopic.php?t=65910&postdays=0&postorder=asc&highlight=stockfish+book&topic_view=flat&start=10

As far as I read the paper, I didn't see a book for Stockfish mentioned, so I guess, match was bookless.

and

http://talkchess.com/forum/viewtopic.php?t=65909&highlight=stockfish+book

While this is indeed incredible, show me how it beats SF dev with good book and syzygy on equal hardware in a 1000 game match.

| improve this answer | |
  • Thanks @SmallChess. Is there a canonical source for this info? I did not find it in the paper. – kuzzooroo Dec 12 '17 at 2:29
  • @kuzzooroo I've added some links. There is no mention at all about opening book for SF in the paper. Please consider to accept my answer. – SmallChess Dec 12 '17 at 4:11
  • 6
    AlphaZero itself is a book, I don't think that the match is fair. – ferit Dec 12 '17 at 6:03
  • 1
    Opening books and tablebases are crutches that a whole generation of chess engine software has grown to become dependent on. The match is fair in the sense that both engines are not relying on pre-recorded moves to play at a higher level, the emphasis is on machine intelligence and not dictation. – prusswan Dec 22 '17 at 9:26
  • 1
    I still haven't found a canonical source for this, but there's discussion all over the place based on Stockfish not having an opening book. – kuzzooroo Jan 28 '18 at 19:06
4

To add to SmallChess's answer, Stockfish as an engine doesn't come with an opening book. Those are separate inputs, and you can give Stockfish whatever opening book you want. Therefore, if the AlphaZero paper doesn't say the authors gave Stockfish an opening book, it didn't have one.

See this thread on Chess.com for more about Stockfish playing with the Cerebellum opening book (aka "Brainfish"). You can see from the rating list that the opening book is worth about 50 elo - a substantial amount. You can also see the difference in the TCEC S14 superfinal, which was played with set openings, and the rapid bonus afterwards, played without books. Leela (an AlphaZero clone) narrowly lost the superfinal, but annihilated Stockfish in the bookless bonus.

| improve this answer | |
  • Leela isn't quiet an AlphaZero clone, IMO. – Brandon_J Mar 21 '19 at 20:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.