Author: Steven Edwards
Date: 15:30:55 08/19/04
Symbolic: Status report 2004.08.19 The new class CTEval has been set up to encapsulate positional evaluation functions that had previously been unnecessarily tied to the toolkit CTCmbPos (combo position) class. The code in CTEval is, unsurprisingly, rather similar to the C code in my old program Spector which in turn was shamelessly cribbed from Chess 4.x by Slate and Atkin. In my defense, I note that S&A lifted much of their evaluation ideas from the 1950 papers by Shannon and Turing. Clearly, chess programming has a long legacy of larceny. The plan is to use CTEval from Symbolic's Lisp source as a tie-breaker to chose from positions that otherwise look the same from the viewpoint of the Lisp exploration code. It's likely that the quick and crude static evaluation functions in CTEval will eventually be replaced by much slower and rather more sensitive code. This future version will be called only from the ChessLisp interpreter and even a relatively long function execution time can be easily absorbed within Symbolic's target node time of 200 milliseconds. From a purist standpoint this future evaluation function should also be written in Lisp, but there is little to be gained from an AI research view in doing so; simple square counting and bitboard operations are nothing new. ------------------- Somehow, a null move bug found its way into the toolkit which allowed a null move to be tried in positions like BCE #70: [D] 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1 This has been fixed and so once again the toolkit picks 1. Kb1 in a nineteen iteration search that takes less time than an eyeblink. ------------------- On the ICS front, a problem with time allocation when using increment controls was detected when Symbolic was playing some G/120+12 games; the program was accumulating too much saved time for no good reason. Also, several losses incurred in recent ICC play due in part to operator error when I mistakenly left a distributed processing task running in the background; it cheerfully stole half the CPU cycles available. Examination of some xboard games against various opponents has shown that Symbolic was too easily trading away its queen for overvalued collections of non queen material. In response I've tweaked the queen/pawn ratio from 9/1 to 933/100. ------------------- The new class CTPost has been added to help support xboard's post/nopost commands and so xboard's "show thinking" menu pick is now operational. Perhaps I'll use the new CTPost capability to assist with sending ICS analysis whispers. ------------------- As more of the toolkit issues are resolved and as time becomes available, I'll be returning to work on Symbolic's Lisp code.
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.