Author: Steven Edwards
Date: 13:07:40 04/25/05
Go up one level in this thread
On April 25, 2005 at 13:37:03, Dann Corbit wrote: >On April 23, 2005 at 06:18:10, Steven Edwards wrote: >>On April 23, 2005 at 01:33:50, Dann Corbit wrote: >>>On April 22, 2005 at 21:21:07, Steven Edwards wrote: >>>What does pes represent? >> >>The EPD mnemonic "pes" means "pawn evaluation symbol" and is a replacement for >>"ce" (centipawn evaluation). It is the expectatation in signed decimal pawns >>with symbols used for special evaluations like mates, losses, even, unknown, >>etc. > >I prefer the old method. A hundredth of a pawn is fine for describing the value >of a position. And if there is a direct mate, then the dm code should be used. >Having to parse new formats for something already adequately described is not >optimal for me. And especially the case when one field has more than one method >of interpretation. Maybe others will like it, though. I can see your point about not wanting to add another EPD opcode. The old "ce" opcode can still be used (of course), but my program won't generate it because the original definition was sligtly buggy and incomplete. (YES, THIS WAS MY FAULT.) 1. The "ce" operand values for mate/lose were an unnecessary exposure of the inner specifics of the old program Spector; they should have stayed hidden and should have not been allowed to create unneeded dependencies in other programs. Not a good thing; I should have done better. 2. The "ce" mate/lose values were not symmetrical and couldn't be symmetric; the opposite of "Mate in five" is not "Lose in five". It could be argued that its opposite is halfway between "Lose in five" and "Lose in four". Actually, half of all the "ce" values in the mating/losing losing range did not even have a well defined interpretation. 3. It isn't the easiest thing for a human to quickly mentally translate "ce" mate/lose values into fullmove distance integers. This applies to humans using grep and the like for scanning. 4. Even if the mate/lose deficiencies are ignored, there is still the problem of representing other special values like "Illegal", "Unknown", and "ForcedDraw" (vs. a non draw balanced position). 5. Centipawn resolution may be good enough for some programs, but it is a reach to say that it is good enough for all programs. Thirty years ago, Chess 4.x was the world champion program and it used only quarter pawn resolution in its ouput. What might the 2035 champion program need? We can't say, so the best bet is to use a floating point value and allow the generating software to select the resolution. This is analogous to the change made in the imaging engines from Mac OS9 (QuickDraw integer coordinates) to Mac OS/X (Quartz floating point coordinates).
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.