Author: Robert Hyatt
Date: 09:14:33 06/29/01
Go up one level in this thread
On June 29, 2001 at 10:47:18, Joshua Lee wrote: > >>It doesn't quite work out like that, which is why everybody switched to a chess >>4.x=type approach during the middle and late 70's. The problem is, if you are >>selective at every ply, you have some probability of making a serious mistake >>by eliminating a critical move. This probability is a multiplicative product >>of the probability of ignoring a critical move at any one ply. The deeper you >>search, the higher the probability that you will simply crash and burn due to >>a tactical error. >> >>Remember, it has never been shown that Nxe6 is a forced win. It is a positional >>gambit that looks good. But since there is no forced win of material within a >>reasonable horizon, this is about positional judgement. Programs back then were >>still reasonable positional evaluators, they were just tactically inferior to >>today's programs due to hardware available and software decisions made because >>of the hardware we had. >> >> >>Belle was based on hardware, even in the middle 70's. It had a hardware >>move generator, hardware evaluation, hardware make/unmake move, software >>search. Chess 4 would require a CDC Cyber 176 as it was written in Compass >>(Cyber assembly language). Those machines are long gone from the computing >>market. >> >>Pioneer _never_ played a game. Never. It was mainly vaporware. >> >>Nuchess still exists. In fact, Dave Slate used to visit ICC quite a bit >>(handle=rusty) and he was still working on it 6-7 years ago, although I don't >>know how much he was doing to it at the time. It certainly played in some of >>the later ACM events, although I don't recall it playing after 1990 or so. > > >Have you thought about a match between Crafty /CB and Nuchess? >What ideas of the Golden Oldie's Aren't in use ? Lots of selectivity. For shallow searches, the selectivity was not bad at all. But the deeper we went, the more errors it tended to produce, because they would enter a critical (forcing) line but had included an error down in the analysis that refuted everything. Programs of the early 80's used null-move, so that isn't really new. My first program understood "the square of the pawn" by 1975 or so, so that isn't new. Everybody had reasonable books (although not nearly as big as today as back then there were no "machine readable" games available unless you typed them yourself.) NuChess would not be an "easy victim" for today's programs. I doubt it would win a match or even come close to it, but it would not "roll over" any more than gnuchess will. >I remember seing a post here about Chess 4.x that it was available at one time >as a PC program in the early 80's does Can anyone elaborate on this? That wasn't 4.x... Slate or Atkin wrote a pascal chess engine called chess 0.5 and had it published over a few months in either Byte magazine or Personal Computing. (Don't remember which, although I probably have the stuff in my files at the office). It was a minimal chess program based on alpha beta, iterative deepening, etc... But made small enough for a PC. I don't even recall if it had hash tables.. > >I plan on looking at some of the older WCCC games to see how Crafty stands up, I >am willing to bet it is only going to have a hard time finding errors/tactics of > Hitech , Deep Thought, Cray Blitz but i am not sure about ply Depth. I had crafty play through the 1986 WCCC games Cray Blitz played. It didn't find any tactical errors given the same amount of time that Cray Blitz used in the match. I believe I posted this on r.g.c.c a few years ago when we were discussing the 1986 event. I was surprised because when I ran the test, Crafty on the quad whatever-I-was-using-at-the-time was faster than Cray Blitz (although not by a huge advantage). Chris Whittington criticized one particular move as "ugly". We later found out that _every_ program would play that move as it was the only way to avoid losing a pawn. His program included... That was how the analysis got started. > >I would like to know what the Opinion of the Programmers here is to what ply >depth is needed to defeat Grandmasters , or /and Play as well as Hitech , Cray >Blitz , DT/DB , Also which programs will be the model of ply? I don't think you can answer that easily. there are "plies" and there are "PLIES". plies with null-move are less effective than plies without them, but since you get more plies with null-move, there is a small gain overall for most. But comparing plies to plies is like comparing body weight to predict how a weightlifter will do. >In Monty Newborn's book "Kasparov versus Deep Blue" on page 238 it says " In >1993 at the twenty-third ACM International Computer Chess Championship, a panel >discussion was held, and it was generally concluded that a fourteen-ply search >by Deep Thought or an equivalent program would be enough to defeat Kasparov. A >thirteen-ply search might, in fact, be sufficient, while a twelve-ply search was >probably not enough." > Deep Thought was a classic full-width program, no null-move, lots of extensions, reasonable evaluation, etc. Except for null-move, it was similar in structure to today's programs. Later versions had significantly more hardware evaluation stuff than we do today, but the basic "design" is not vastly different. A 14 ply search without null-move is a big search. As seen by the speed of the hardware deep blue two required to reach this number and beyond. A 14 ply search with null-move is a big search in the middlegame, in fact. >I would like to know what exactly is an "Equivalent Program" > What's the criteria??? >What did Deep Thought Use to make it so different and what program now should be >the example? To approximate deep thought, you _could_ use crafty. Here is what you would have to do: 1. set selective=0/0 to turn null-move totally off. 2. Add Hsu's singular extensions code. 3. run it on hardware fast enough to search 5-10M nodes per second. That would be tactically similar. The evaluations are probably significantly different, so positionally they would not be very similar... DB is another question altogether of course... much faster. Much larger evaluation. Mysterious pruning in the chess processors. two-level parallel search. etc. > >Thankyou
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.