Author: Bryan Hofmann
Date: 10:34:25 11/10/03
Go up one level in this thread
On November 10, 2003 at 13:23:31, Robert Hyatt wrote: >On November 10, 2003 at 12:52:57, Bryan Hofmann wrote: > >>On November 10, 2003 at 12:36:46, Robert Hyatt wrote: >> >>>On November 10, 2003 at 11:33:17, Bryan Hofmann wrote: >>> >>>>A few questions as I have seached high and low and have not found any answers to >>>>these; >>>> >>>>1. When preforming a "make profile" (with gcc) do you use the book.bin and >>>>books.bin files? >>> >>>Yes, although there is no need to do so. I have one book position in the >>>profile positions to produce profile information for the book selection code. >>>Why, I don't know, since this is not exactly time-critical code. :) >> >>I'm not sure I follow you with the one book position... I compiled crafty the >>first time without the profile and created a book.bin from you gm2600.pgn file. >>Should I be using this book.bin with the make profile? Also how long does it >>take to compile with the profile (it has been running now for more then 2 hours >>on a quad 450mhz, it is running the last runprof command and is at depth 15 even >>tho the st=10)??? >> > >Which book you use really doesn't matter. The point of the profile run is >to simply execute most of the code so that the branch probabilities can be >discovered and used to further optimize the executable on the second build. > >make sure you have a "ponder=off" in your .craftyrc/crafty.rc file. If >not, the test will hang as you are seeing. Thanks, I was using the standard Makefile with no .craftyrc file and it did not have ponder=off in the last runprof. > > >> >>> >>>> >>>>2. How can you make Crafty learn from a EPD after it has performed an analysis? >>>>(ie import into the book.lrn). >>> >>>I am not sure what you are asking about. When crafty plays a game, it >>>automatically learns and updates the book.bin file. the book.lrn file is >>>just an ascii representation of what it learned that can be given to other >>>Crafty users, or whatever. You can use the "import" command but there is >>>no need as it has already been included in the book.bin information. >>> >> >>I downloaded a CCC.EPD file that was created that lead to mating. If I run the >>epdpfga <infile> <outfile> I get the analysis but there is no change to the >>book.bin nor the book.lrn. >>> > >Those don't affect the book, as they are not book positions... > > >>> >>> >>>> >>>>3. When creating a book file (from the gm2600.pgn on Dr. Hyatt's site) is is >>>>better to do a 60 2 100 or a 60 2 50 for stronger performance? >>>> >>> >>>Here is the issue. For every move played in a game, there is a probability >>>that it is either not best, or even that it is an outright blunder made near >>>a time control. If you limit the book lines to something pretty short, you >>>tend to reduce the probability that your "game" has an error in it, for >>>obvious reasons. The shorter, the less chance there is a gross blunder. Of >>>course, if you reduce the length to zero you are _certain_ there are no >>>blunders. :) >>> >>>I tend to use 50 as the normal number for my books. But it is probably >>>irrelevant if you use the "played N times" value. IE here is the way I >>>build the book I use on ICC: >>> >>>book create file.pgn 60 10 50 >>> >>>This says create a file with each line being no longer than 60 plies, >>>each move played at least 10 times (at least 10 games have this move >>>in this position played) and at least 50% of the games were won by the >>>side to move. >>> >>>That makes the book fairly reliable, and also fairly small. >>> >>>That's how the book.bin file on my ftp machine was made from the >>>enormous.pgn file also located there. This produces a 13 mb book >>>file which is managable and pretty reliable. >>> >> >>Great!! This is the info I was looking for... Also thanks for creating a great >>chess engine with source code available to all! >> >> >>Bryan >> >>> >>> >>> >>>> >>>>Thanks >>>> >>>>Bryan
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.