Author: Matthew Hull
Date: 15:19:28 12/01/04
Go up one level in this thread
On December 01, 2004 at 17:54:21, Uri Blass wrote:
>On December 01, 2004 at 17:20:16, Matthew Hull wrote:
>
>>On December 01, 2004 at 16:33:50, Peter Berger wrote:
>>
>>>On December 01, 2004 at 16:11:14, Roger D Davis wrote:
>>>
>>>>
>>>>>
>>>>>Movei played on Athlon64 3200+(2.0GHz), 64MB Hash, Junior on a P233MMX, 16MB
>>>>>Hash - looking forward to the next games.
>>>>>
>>>>
>>>>You're telling me that Movei was running on a computer about 8-10 times faster
>>>>than Junior?
>>>>
>>>>Roger
>>>
>>>Hi Roger,
>>>
>>>I guess it is more like 12-13 times faster, but that's OK - remember the
>>>Crafty-Junior match when Crafty only ran on a PIV 2.0GHz and barely managed to
>>>win by 6-5.
>>
>>
>>What would the score have been with Crafty using the WCCC04 Berger book?
>>
>>:)
>>
>>I guess books can make some difference, yes? The professionaly made commercial
>>book against the Crafty book? How many points is that worth?
>
>I do not think that the professional prepare special lines against Crafty's
>book.
Right. But the commercial book is tuned for Junior by a human, presumably,
specifically for Junior's playing strengths.
Crafty's book is not the product of that level of preparation.
Yes?
In other words, Peter's book at WCCC for Crafty is a better book. Right?
>My opinion is that book is overestimated and if the professional wanted to
>prepare special books against Crafty then you could see them often get
>evaluation of more than +1 when they are out of book and it does not happen.
>
>about books it is possible that movei's book is too short but I prefered to do
>the mistake of too small book than the opposite book.
>
>Movei's book is based on games of top players that I downloaded from the
>internet.
>
>White's book is based only on wins when black's book is based on draws and wins
>so black's book is bigger.
>
>minimal frequency for book move is 5 times so with white you need to win at
>least 5 games in order to include the line and with black you need not to lose
>at least 5 games to include the move in book.
>
>I plan to add in the future possibilities to add moves to the book and to delete
>moves from the book by special add and delete book so people who will like to
>build better book for movei will have better tools to do it and will not need to
>delete games in order to do it.
>
>last note about movei's book.
>
>Movei can build a book from every pgn that has at most 9,000,000 moves in the
>relevant plies(I am not sure about bigger pgn and if the number of book moves
>before compressing is more than 9,000,000 there may be some overflow problem)
>
>The pgn that was used has slightly less than 50,000 games in total and only 40
>plies were used as maximal ply of book and 50,000*40=2,000,000 so there was no
>problem with it.
>
>relevant line from movei's code is
>bbuffer=(book_first_info *) calloc(MAX_SIZE_BOOK,sizeof(book_first_info));
>
>typedef struct tag_book_first_info
>{
> unsigned char position[8];
> unsigned char move[2];
> int frequency;
> int learning_value;
>} book_first_info;
>
>You can see that 9,000,000 moves needs 9,000,000*18 bytes=162,000,000 bytes so
>it is possible that people with less than 162,000,000 bytes of ram will be
>unable to build a book for movei but I guess that today most users have more
>than it.
>
>Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.