Author: Vincent Diepeveen
Date: 10:23:45 01/18/00
Go up one level in this thread
On January 17, 2000 at 16:46:52, Rajen Gupta wrote: >I remember when i 1st started reading the web comp chess reports-rebel8 was the >new champ-by a huge margin.since then there is no convincing evidence that later >versions of rebel are stronger than rebel8. perhaps someone who does extensive >testing like Mark Young or enrique can tell us (on the basis of actual tests and >comp vs comp games please, not merely on subjective impressions)the following: About Rebel. First of all let's see what did Rebel get so high at SSDF, as i guess that's what you refer to? Some months before rebel8 came out i emailed with Ed Schroeder. I emailed asking him about whether lazy evaluation worked for him. Ed denied using lazy evaluation. Some months later Rebel8 came out, basically searching a lot of nodes a second faster than rebel7/6, apart from that i didn't have the feeling rebel8 was much different from 7. Some say it was positionally weaker than 6/7. Well exactly that happens when using lazy evaluation. Recently Ed said he *always* used lazy evaluation in Rebel. Ed probably already somewhere in rebel used lazy evaluation. I felt rebel8 was tactical anything but weak. For SSDF however rebel8 had 2 new things. First of all a big tournament book from which each line was already auto232 tested. Secondly, and this gets really underestimated by everyone, it aborted games that were the same, within 2 moves out of book. Now obvious i'm not a fan of playing the same game over and over again, but considering the nature of the book in rebel, which has some lengthy and wide lines which i call 'killerlines' (lines that objectively aren't representing the state of the art theorem, but where you know in advance that you win against certain other programs with, as they 'fall' for the line). So if you win 20 games from rebel8 with 1.d4 ... 2.a3 then in fact your games get 19 time aborted Yet if in the richter rauzer a certain Qxe5 side line wins for rebel, then you might lose 10 games in a row, as in a positoin where you're already dead lost, there rebel is still in book having several possibilities. Further the interpretation of the games. Rebel finds in endgame pawns worth very little. Let's look to rebel: boring openingsbook, but very good book. No questions about that. This means that a game rebel wins is usual SHORT. A game that it loses it doesn't get bad out of book usual, so that is usual a rather long win. Auto232 player aborts such a game after a certain amount of moves, then rebel as it evaluates dead lost positions usually under -5, it puts a '?' so a question mark as the result of the game. I forgot but don't aborted games which are repeated twice are also carrying a '?' as result? Anyway, human factor gets *heavily* underestimated when interpreting results. I'm not sure what SSDF does, how many '?' results does Karlsson receive? Anyway, this is already enough to explain the rating jump of Rebel at SSDF. >which is the strongest version of rebel >is rebel getting progressively weaker with each version or is this merely >because other programmes are getting disproportionately stronger? Of course a program never gets weaker. Others just learn how to beat it. >if rebel is actually getting weaker then what is the purpose of releasing new >versions? Don't you want to get updated with new versions? >rajen
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.