Author: martin fierz
Date: 13:42:47 07/16/02
Go up one level in this thread
On July 16, 2002 at 16:20:30, Russell Reagan wrote: >On July 16, 2002 at 15:14:17, Omid David wrote: > >>Each program has its own style and thus should play openings which fit its >>style. Of course these openings should be tuned by the programmer, so a good >>deal of chess knowledge is necessary on programmer's behalf. >> >>As an example, in tournaments I always play Sicilian against 1.e4 and adopt the >>Scheveningen variation, so I need to have a vast knowledge of Scheveningen (and >>Najdorf) while I can forget about Caro-Kann! Programs too should have their own >>opening repertoire consisting of an extensive knowledge of all variations which >>might occur in their chosen opening variations, while they can forgo other >>variations. >> >>Omid. > >Thanks Omid. I have another question. Let's take an example. Let's say that my >program has a Ruy Lopez line that goes to move 20, and then the book stops >there. Would it be advantageous to let my program think from that position for, >say, a week, and then add that move to my book? That way I have a week's worth >of search instantly. > >The only thing that I forsee as a problem is the same problem that a beginner >who memorizes book lines has. For example, let's say my program plays the last >move in the book (the one that was searched on for a week), and it plays it, and >my opponent responds, and then my program doesn't have access to the long >variation that it had in the week long search, so now my program *might* (not >sure about this) be just like the booked up beginner who doesn't know what to do >when it leaves the book. For example, the program might need to search to 20 >plies to "see" the reasoning for it's previous move, and if it can only get to, >say, 16 plies in the given time slot in a game situation, it might not follow >that PV line. Do you see this as something that might be a problem? > >Thanks, >Russell yes, what you describe is certainly a problem. i generate the opening book for my computer checkers program automatically, with the engine calculating positions and adding them to a database. i usually just use a tournament time control level to search a move, because doing shallower searches can hurt you because the book move is worse than what you would see on a normal search, and doing deeper searches risks not finding the necessary follow up. it makes much more sense to expand your line with something like 240 5-minute searches instead of 1 12-hour search. you get many more book moves, and since you back up scores from later searches to the root, you have a similar kind of quality for the first move in your computed book as if you had searched for 12 hours there. aloha martin
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.