Author: Robert Hyatt
Date: 21:34:45 10/10/98
Go up one level in this thread
On October 10, 1998 at 12:28:54, Johanes Suhardjo wrote: >On October 10, 1998 at 11:17:54, Robert Hyatt wrote: >>let me clear those up more. "no tricks" is a flag that turns off a dangerous >>piece of code in the eval that is used to handle the "trojan horse attack". >>But it also makes crafty paranoid about an open h-file with queens and rooks >>on the board. I view this as unsafe and only turn it on against players >>known to use the trojan horse (this is automatic in crafty, and uses the >>infamous "S" (special player) list. I generally have this "on" (no tricks >>on which means don't use the tricky eval stuff). > >I hate the idea to have special code for special opponents. What if other >players adopt those special tricks? parrot had two "customers" on FICS that >beat it consistently (one of them was about 400 points below him!!). So I >wrote special code to handle their tricks. Worked like a charm, but since I >activated the code for everybody, it made parrot play weaker in general. > > >>the "draw score normal" should not be used vs a computer. What this >>does is say that drawscore=-.66 in the opening, -.33 in the middlegame >>and 0 in the endgame, but it is modified when the opponent is in time >>trouble. If this is "false" (0) then draw score is set to 0 and left >>there, which is the *only* way to play computers, unless I *know* I am >>much stronger than the opponent. This is dangerous, but helps against >>humans that are rated lower and just want to draw over and over. It >>fights like hell to avoid draws. :) > >Yeah, recently I added that rule (after reading one of your CCC articles). >I needed it because there are several human opponents that took advantage >of their lower rating by moving pieces back and forth. Either the games >ended in a draw or parrot sacrificed pawns to avoid the draw and ended up >in a losing position. In any case, after I added that rule, parrot scores >better now against humans. It scores worse againts computers though. I >thought that was because other programs improved, but now maybe I have to >think again. > >Please note that rating means nothing to me, but I need a guide of whether >my programming is going to the right direction. So I limit FICS rated games >to 200 points above and below parrot's rating. That seems to be more >reliable. No more much lower human opponents trying to get quick draws, no >more that dangerous one-trick-parrot-kiler. 8-) > > here's why a "dynamic draw score" won't work against computers. I used this approach for a good while and noticed that suddenly my results against a couple of computers dropped, "lonnie" being one of them. I was using the thing that says draw=-.6 in opening, -.3 in middlegame, and 0 in the endgame... What lonnie did was set his "contempt factor" to +.3, which means that his "engines" would seek to repeat when possible... and crafty would be happy dropping to -.3 to -.6 to avoid repeating. And such a poor position would let it avoid the draw and instead lose outright later. All because a human noticed what I was doing and exploited it. So now, against computers, draw=0, regardless... and I don't hve that problem. I do draw=0 in endings because even against a human I don't want to wreck my position to avoid a draw, because I may avoid it by losing... > Johanes Suhardjo (johanes@farida.cc.nd.edu) >-- >Count your garden by flowers, >Never by the leaves that fall. >Count your days by golden hours, >Don't remember the clouds at all. >Count your nights by stars, not shadows, >Count your life with smiles, not tears, >And with joy on every birthday >Count your age by friends, not years. > >Italian Philosopher
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.