Re: [computer-go] want zobrist's thesis

[Vincent Diepeveen <diep@xxxxxxxxx>]

Not at all,
but you might want to read for example something from Dennis Breuker about
hashtables in computerchess.

He is clearly very busy with Zobrist hashing there and has formula's to
estimate the chance for both collission and errors.

At supercomputer with 200 processors i recalculated the chances and all the
theoretic publications with 115GB hashtable is complete BS.

In Go i advice 128 bits Zobrist hashing. Then rest of your life you don't
need to worry about it when cpu power gets 1000x faster and memory capacity
100 times.

You can for example store 64 in hashtable and use part of the other 64 bits
as an index to the hashtable. Either with % or with AND or with some
multiplication trick.

All theories are just a big nonsense in Zobrist because
  a) there is no perfect random numbers
  b) while playing a game the search space non stop is changing depending
upon how deep you search (if not selectively).

In Go the search space is bigger so you need more bits.

In DIEP i'm storing 51 bits in hashtable and use the other 64 bits (part of
it) as a lookup in hashtable.

Vincent

At 17:12 1-11-2004 +0800, zhaohui yin wrote:
>does anyone had A.zobrist's thesis "A model of visual organization for
>the game of Go", and the document about zobrist hash ?
>
>yinzhaohui
>_______________________________________________
>computer-go mailing list
>computer-go@xxxxxxxxxxxxxxxxx
>http://www.computer-go.org/mailman/listinfo/computer-go/
>
>
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/