Author: Dann Corbit
Date: 21:40:15 12/08/05
Go up one level in this thread
On December 08, 2005 at 23:57:51, Uri Blass wrote: >On December 08, 2005 at 23:35:36, Dann Corbit wrote: > >>On December 08, 2005 at 23:19:39, Uri Blass wrote: >> >>>On December 08, 2005 at 19:32:53, Dann Corbit wrote: >>> >>>>On December 08, 2005 at 19:09:42, Uri Blass wrote: >>>> >>>>>On December 08, 2005 at 18:27:01, Claude Le Page wrote: >>>>> >>>>>>Sevral posts have noticed that Rybka is rather weak in endgames ,especially >>>>>>against Junior : this put me on apossible track : when its position is not >>>>>>clearly better , a Juhior engine uses its fastness to complicate , and Rybka >>>>>>must follow at maximum depth ,what is very costly in time : so , in endgame , >>>>>>Rybka is in zeitnot , and analyzes only at depth 13 where thr other is at depth >>>>>>17 >>>>>>The same happens between Junior and Hiarcs : Their Knowledge slowen hem too >>>>>>much in front of engines that use long calculated lines >>>>>>The same seem to happen vs TogaII 1.1a , whose style is similar to Juniors >>>>>>Could this be a possible explanation? >>>>> >>>>>The reason that rybka has problems in endgame is simply lack of knowledge about >>>>>the endgame. >>>>> >>>>>I do not understand why people try to find other excuses. >>>> >>>>Perhaps Rybka is a program that could benefit greatly from EGTB (such that a >>>>clear increase in Elo does occur from their use). >>> >>>I think that it can earn more from endgame knowledge. >> >>It is an interesting idea to me to try it both ways. I am thinking about it >>this way: >> >>Adding endgame knowledge will slow down the eval. > >It is not a problem for rybka. > >As Vasik explained in the readme file > >"Rybka aims to have a fully knowledgeable evaluation function. This term however >has taken some abuse recently, so let me make something clear: chess knowledge >wins chess games. If it doesn't, it isn't knowledge." > >It already has a lot of middlegame knowkledge in the evaluation and it is >productive and it is clear that adding endgame knowledge is an easier task as >Vasik said in one of his posts. > > > Perhaps leaving it out and >>allowing the EGTB to handle endgame is good enough. But what happens when you >>have not got all the way to EGTB land (e.g. 10 chessmen on the board). > >You cannot expect more than 7 piece tablabases in the next few years so I do not >think that it is relevant. If there are ten chessmen on the board, then 4 captures means 6 men hits. Crafty sometimes probes endgames when lots of chessmen are left on the board. For instance, with 14 chessmen and 5 man tablebase files, I got ten probes here: [D]3bkb2/3pnp2/4q3/8/8/4Q3/3PNP2/3BKB2 w - - White(1): epdpfga \probetest.epd probetest.out PFGA: EPD record: 1 clearing hash tables time surplus 0.00 time limit 16:39 (+0.00) (16:39) depth time score variation (1) 8 0.84 0.13 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Bb6 4. Ba4 f5 5. Kd3 8 1.17 0.27 1. Bb3 Qf6 2. Bg2 Qa1+ 3. Bd1 Bg7 4. Be4 Ba5 5. Nf4 8-> 1.44 0.27 1. Bb3 Qf6 2. Bg2 Qa1+ 3. Bd1 Bg7 4. Be4 Ba5 5. Nf4 9 1.67 0.10 1. Bb3 Qf6 2. Bg2 Qa1+ 3. Bd1 Bg7 4. Be4 Bc7 5. Nf4 Bd4 9 2.25 0.11 1. Bh3 Qc6 2. Bb3 Bg7 3. d4 Ba5+ 4. Kd1 Qh1+ 5. Kc2 Qc6+ 6. Nc3 d5 9-> 3.78 0.11 1. Bh3 Qc6 2. Bb3 Bg7 3. d4 Ba5+ 4. Kd1 Qh1+ 5. Kc2 Qc6+ 6. Nc3 d5 10 5.77 0.15 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Bf1 Bh6 4. Qe5 Kf8 5. d3 Ng6 6. Qe4 10-> 8.17 0.15 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Bf1 Bh6 4. Qe5 Kf8 5. d3 Ng6 6. Qe4 11 38.69 0.26 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qa8 4. Bc2 Bb6 5. Qe5 Qa7 6. Nb5 Qb7 11-> 42.52 0.26 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qa8 4. Bc2 Bb6 5. Qe5 Qa7 6. Nb5 Qb7 12 1:03 0.27 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qd5 4. Bb3 Qh5+ 5. Qf3 Qxf3+ 6. Kxf3 Bb6 7. Nb5 Nc6 8. Bf5 Bg7 12-> 1:17 0.27 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qd5 4. Bb3 Qh5+ 5. Qf3 Qxf3+ 6. Kxf3 Bb6 7. Nb5 Nc6 8. Bf5 Bg7 13 2:18 0.34 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qa8 4. Nb5 Qd5 5. Ba4 Qh5+ 6. Kf1 Bh6 7. Nd6+ Kf8 8. Qd3 Qd5 9. Qxd5 Nxd5 10. d4 13-> 2:49 0.34 1. Bh3 Qc6 2. Nd4 Qh1+ 3. Ke2 Qa8 4. Nb5 Qd5 5. Ba4 Qh5+ 6. Kf1 Bh6 7. Nd6+ Kf8 8. Qd3 Qd5 9. Qxd5 Nxd5 10. d4 14 5:56 0.40 1. Bh3 Qxe3 2. dxe3 Ba5+ 3. Kf1 Bg7 4. Ba4 Nc6 5. Kg2 Ke7 6. Bf5 Bb6 7. Kg3 <HT> 14-> 7:34 0.40 1. Bh3 Qxe3 2. dxe3 Ba5+ 3. Kf1 Bg7 4. Ba4 Nc6 5. Kg2 Ke7 6. Bf5 Bb6 7. Kg3 <HT> 15 12:09 0.41 1. Bh3 Qxe3 2. dxe3 Ba5+ 3. Kf1 d5 4. Ba4+ Kd8 5. Bb3 Bh6 6. Ng3 Bb6 7. Ke2 Kc7 8. Nf5 Nxf5 9. Bxf5 Kc6 10. Kf3 time=16:39 mat=0 n=1230558683 fh=92% nps=1.23M ext-> chk=91056010 cap=3089816 1rep=4459212 mate=190814 predicted=0 nodes=1230558683 evals=870169208 50move=0 endgame tablebase-> probes=10 hits=10 SMP-> split=0 stop=0 data=0/32 elap=16:39 White(1):
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.