Author: Tony Werten
Date: 23:53:01 10/29/01
Go up one level in this thread
On October 29, 2001 at 11:54:18, Robert Hyatt wrote: >On October 29, 2001 at 10:43:35, Tony Werten wrote: > >>On October 29, 2001 at 10:29:26, Robert Hyatt wrote: >> >>>On October 29, 2001 at 04:05:08, Tony Werten wrote: >>> >>>>On October 28, 2001 at 21:06:11, Robert Hyatt wrote: >>>> >>>>>On October 28, 2001 at 18:20:43, Bas Hamstra wrote: >>>>> >>>>>>Again many bizar events. While discussing how boring and anti-positional Tiger >>>>>>usually plays, with Jeroen Noomen, during the game Tao-CT, CT shot a h-pawn >>>>>>through Tao's left eye. End of discussion. Well, let's call the thing an >>>>>>idiot-savant, that's as far as I can go. Newcomer is EEC that does not use any >>>>>>form of AlphaBeta at all. The author tried to explain the concept to me, but I'm >>>>>>not sure I understand. Each move it plays a couple of shallow "matches" against >>>>>>itsself and some kind of pattern matching is involved. Eventually that leads to >>>>>>"votes" for different moves. The author says it has won some Blitz games against >>>>>>GNU. >>>>>> >>>>>>Also new is "Gadget" from Hans Zijdenbos, written in Basic. Since it is brand >>>>>>new, it has many things *not*. It has no hashtable, no pondering, no nullmove >>>>>>and believe it or not, no quiescence search. Very little evaluation. Now can >>>>>>that play chess? Amazingly, it does. Imagine Gadget searching 6-7 ply and Tao >>>>>>12-13 ply, basically twice as deep. Yet it kept Tao negative for over twenty >>>>>>moves, with sound play. Cock de Gorter said the same happened during >>>>>>Crafty-Gadget. You would almost expect non nullmovers find positional holes in >>>>>>the nullmove search or something. >>>>>> >>>>>>During a deep technical discussion with Johan de Koning whether to analyze mates >>>>>>and countermates in the evaluation function or not, Jeroen Noomen drops by: >>>>>>"aha, a rabbit". Apparently an enemy knight at h7, that can't move, is a rabbit. >>>>>>We didn't get it, what do rabbits usually do? We could think of two things: a) >>>>>>multiply and b) disappear in a hole. Since there were no minor promotions >>>>>>possible in the foreseeable future and the rabbit finally disappeared in h7 I >>>>>>wouldn't be surprised if b) has something to do with it. >>>>>> >>>>>>Well, Tao scored 3 out of 6 so far, on a PIII-500. Should have won against Ant, >>>>>>which was positionally manoevred into a corner quite nicely, but alas, fell into >>>>>>the "impotent pair" trap which I have no code for: bishop+unpromotable pawn, and >>>>>>it became a draw. >>>>>> >>>>>>Finally: what's the matter with Crafty? It has the fastest hardware of all, but >>>>>>seems to do relatively poorly, what's the matter? The new SE stuff, or what? >>>>>> >>>>>>TooTheLoo, >>>>>>Bas. >>>>> >>>>>1. Does Crafty _really_ have the fastest hardware there? That's hard to >>>>>believe with lots of 1.4 gig Dual AMD machines cheaply available. But in any >>>>>case, it is possible. A dual 1.4 is faster than a quad-700 for Crafty, however, >>>>>as the quad loses a bit more to overhead... >>>> >>>>Crafty doesn't have the fastest hardware but it's still quite fast. In the >>>>endgame Crafty - XiniX I was outsearching Crafty several times by 3 to 5 ply on >>>>a celeron 700 ! Unfortunately it wasn't enough to win, but it does seem to >>>>indicate something is wrong in Crafty's endgame. XiniX had no problem searching >>>>the same depth. >>>> >>>>Tony >>>> >>>>> >>> >>> >>>I don't think the bug shows up like that. If there is a bug at all. Are you >>>using the 3-4-5 piece EGTBs? That definitely affects the search speed. >>>Otherwise, no ideas here... >> >>Yes. I probe 3-4-5 tables from disk in the search (not in quiescence ) XiniX was >>not faster than Crafty in nps (but 4 times slower). In the positions the >>branchingfactor is so small that 20 plies should not be a problem. >> >>Tony > > > >Can you post the position (or if you already have, indicate where) and I will >look. One easy explanation is that if you see a score of (say) 3.00, while >Crafty is saying 0.00, then the two programs are searching trees of vastly >different 'shapes'. IE you can prune every 0.00 type branch while Crafty >can't. Don't have the possibility to check now. IIRC there were 2 positions were Crafty said 0.00 that are doubtfull. The first one is when Crafty exchanged it's rook for a bishop (on g2?) and a pawn. Second (this is the one were the big search differences came ) is right before or after Crafty exchanged it's bishop for a knight (on c3?) and gave up on the bishoppair. The zero score might be correct here, but I don't think an engine should be able to see it. Tony > >This is often the reason for dramatically different search depths, when one >program has an evaluation closer to the truth than the other. Or at least >when one program has an evaluation that greatly enhances the pruning while >the other has an evaluation that is forcing it to consider lots of lines >that are not needed. > >> >>> >>> >>> >>>>>2. Crafty may well have a serious bug. Michel has reported that it will crash >>>>>on a deep think, which is odd. There is no SE code in it, but it is possible >>>>>that something in recent versions has left a bad array subscript or something. >>>>>I am trying to get it to repeat a crash, but so far, nothing... >>>>> >>>>> y
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.