Author: Vine Smith
Date: 14:21:22 04/09/01
Go up one level in this thread
On April 09, 2001 at 09:11:58, Robert Hyatt wrote: >On April 09, 2001 at 01:35:12, Vine Smith wrote: > >>On April 08, 2001 at 22:37:25, Robert Hyatt wrote: >> >>>On April 08, 2001 at 05:16:40, Vine Smith wrote: >>> >>>>So I get the impression that when I try adjusting the tropism, king safety >>>>factors, etc., I'm doomed to disappointment, I won't get "immensely aggressive" >>>>Crafty, so much as "immensely unsound" Crafty. Well, I'll take a look anyway. I >>>>just had ver. 18.9 analyze a game where an IM did a very bad thing to me at the >>>>National Open this year -- Crafty (along with every other program I've had look >>>>at this, including Fritz 6 and Century 3) doesn't understand what's going to >>>>happen until it's way too late. But perhaps the vicious attack my opponent >>>>launched will look more interesting to Crafty with heightened tropism, and a >>>>higher bonus for reducing my king's safety -- it's too deep to spot through >>>>normal analysis. >>> >>> >>>The problem will turn into balancing the aggressiveness vs its normal positional >>>skills. It has a pretty good concept of pawn structure, endgames, etc. If the >>>king safety aggressiveness overwhelms these terms, then you will get some >>>aggressive play that leaves it in dead lost positions when the attack doesn't >>>work. >> >>Well, that sometimes happened to Tal. But whose games are more entertaining, >>Tal's or Petrosian's (yes, I know, some will indeed say Petrosian)? However, so >>far I haven't got anywhere with tropism 125-150, asymmetry 25-50, so I guess >>it's time to ratchet up king safety itself as well. How small are these >>positional values versus material values? I'm trying to get the program to spot >>an exchange sacrifice based on general principles alone, because there's no way >>it's going to search deep enough to see the concrete justification. If it >>sometimes incorrectly sacrifices as well, so be it. Human players carelessly >>toss material around all the time, it's time programs join them! > >The scores can get pretty large. IE king safety can go way beyond a pawn, >and when you factor in the second order stuff, it can go beyond 2 pawns. The >default for asymmetry is -30 to -40 depending on the version. +40 to +50 >should begin to make a difference. > >When you change one of those it should display a vector of numbers showing how >the score will climb or fall as the feature becomes more pronounced or >significant. I did observe the table of adjusted factors, but I wasn't sure how these related to the value of material factors, or other positional factors. I think I need to drop the idea of analyzing this one game, and observe actual play in a variety of positions instead.
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.