Author: Robert Hyatt
Date: 06:35:22 02/22/00
Go up one level in this thread
On February 21, 2000 at 21:46:09, Christophe Theron wrote: >On February 21, 2000 at 21:33:23, Robert Hyatt wrote: > >>On February 21, 2000 at 21:03:19, Christophe Theron wrote: >> >>>On February 21, 2000 at 16:52:03, Robert Hyatt wrote: >>> >>>>On February 21, 2000 at 13:33:58, Christophe Theron wrote: >>>> >>>>>On February 21, 2000 at 12:16:49, Tom Kerrigan wrote: >>>>> >>>>>>On February 21, 2000 at 11:37:52, Mark Taylor wrote: >>>>>> >>>>>>>An idea I had was to have a small incrmental value subtracted from the eval, >>>>>>>this small increment getting larger the deeper into the tree search the eval was >>>>>>>returned from. I had already done this for the values WON & LOST, but I >>>>>> >>>>>>This is a good idea. However, most chess programs have transposition tables. The >>>>>>ideas are not compatible, because ttables assume that a position's score is >>>>>>constant. You will probably want to have ttables instead of your penalty, >>>>>>because once in a while there are huge benefits to having a ttable. >>>>>> >>>>>>>What I did in the end I made the first search iteration look at positional eval >>>>>>>& material eval, then subsequent iterations looked at material eval only - but >>>>>>>this was really a cop out. >>>>>> >>>>>>Yeah, I think this just confuses things. A long time ago I think there was a >>>>>>program that ran on two CPUs. One CPU ran the regular evaluation function and >>>>>>one was material-only. They checked each other. But programs these days get >>>>>>along fine without material-only eval. >>>>>> >>>>>>-Tom >>>>> >>>>> >>>>>The very old program Tech (I think the author was Gillogly, correct my spelling >>>>>please, it was back in 1960) did this, but on only one processor I think. >>>>> >>>>>It played rather well, but was seriously handicaped by lack of deep positional >>>>>understanding. >>>>> >>>>> >>>>> Christophe >>>> >>>> >>>>No. Tech was a 1970+ program. >>> >>> >>>In the book "Echecs & Mips" written by Frédéric Louguet I have found: "1960: >>>Creation of the Tech program, first program able to compute complicated tactical >>>positions". >>> >>>I remember from articles written by David Levy that Tech used a deep tactical >>>search (whetever deep meant at that time), but that positional evaluation was >>>applied to the root moves only. >>> >>>I think this approach could be used by beginners in chess programming. The >>>program could be surprisingly strong (compared to beginner's usual programs). >>> >>> >>> Christophe >> >> >>that is correct. I thought you were talking about using two processors, one >>for the tactical search and one for the normal search (ie like Sun Phoenix did). > > >No actually I missed the point in Tom's post. I did not know about Phoenix. > > > > >>Tech was all about being fast, at the expense of making gross positional errors >>deep in the search (if you can call 3-4-5 plies 'deep'. :) > > >Do you remember if he was doing a QSearch, or just used a SEE to terminate the >search? > > > Christophe No I don't, but Jim and I have corresponded regularly over the past 30 years, and I don't mind asking. What would be fun would be to see if he still has source code... :) Just to see how it would play on state-of-the-art hardware. > > > > >>>> The program that did two searches was called >>>>"Phoenix" by Jonathan Schaeffer. He ran a normal search with several >>>>workstations in parallel, and a "minix" search using several more workstations >>>>in parallel. Minix searched deeper looking only for tactical refutations of >>>>the moves being considered by the normal search. >>>> >>>>Tech was a very fast, very "dumb" type technology approach, which is where >>>>its name came from (tech). Jim occasionally posts on r.g.c.c and can be >>>>reached there.
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.