Author: Bernhard Bauer
Date: 06:49:11 03/09/00
Go up one level in this thread
On March 09, 2000 at 08:32:03, Andreas Stabel wrote: >On March 09, 2000 at 07:31:15, Bernhard Bauer wrote: > >>On March 08, 2000 at 20:19:48, Bruce Moreland wrote: >> >>>On March 08, 2000 at 16:10:18, Côme wrote: >>> >>>>On March 08, 2000 at 15:41:45, Bruce Moreland wrote: >>> >>>>>Are you sure that 1. ... c5 is the best continuation? What about 1. ... Bf6 2. >>>>>Be5 Kg7? >>> >>>>After 1. ... Bf6 ; Be5 and Kg7, white IMHO can play Bxf6+ and White position is >>>>probably winning because he has a pawn up and the black king is misplaced. >>>>Have you try this position on ferret ? >>> >>>No, that's its plan after thinking for 24 hours on this on a crusty old machine. >>> The move might be bad, but just by looking at it with my eyeball it's an >>>interesting alternative to 1. ... c5. >>> >>>You are right that white is a pawn up and is better, but the implication behind >>>my program choosing this continuation is that it has figured out how to do the >>>same or better against 1. ... c5. Its score is +1.35. >>> >>>I think this is an interesting problem, but very prone to misinterpretation. >>>There are two aspects: >>> >>>1) Finding the move in the first place, which involves discovering that the rook >>>can't be taken, and preferring the resulting position. >>> >>>2) Deciding that in the ensuing tactics, white has significant material >>>advantage or a forced mate. >>> >>>You've gotten a lot of replies from people who have found programs that can get >>>as far as the first aspect of this, and want to stop there, satisfied, but so >>>far nobody has found a convincing crush. Are you sure that one exists? >>> >>>bruce >> >>Here is what Crafty did on a 2xP3-450 with modified (lowered) extensions. >> >> (4) 16-> 77:05 0.00 1. Rc4 Rd7 2. Kg2 Qd8 3. Rh1 Bd6 4. >> Rc1 Be7 >> (3) 17 87:17 -0.04 1. Rc4 Rd7 2. Kg2 Qd8 3. Rh1 Bd6 4. >> Rc1 Nh7 5. Re1 Nf6 6. Kg1 Bxg3 7. fxg3 >> Qc7 8. Qf2 Rcd8 9. Rec1 Nd5 >> 17 207:41 ++ 1. Rxh5!! >> 17 507:30 0.67 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qb4 9. Rxg6+ Kh8 >> 10. h5 Kh7 >> 17-> 509:15 0.67 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qb4 9. Rxg6+ Kh8 >> 10. h5 Kh7 >> 18 611:06 0.79 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qd1+ 9. Kg2 Qd6 10. >> Rxg6+ Kh8 11. h5 Kh7 >> 18-> 622:38 0.79 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qd1+ 9. Kg2 Qd6 10. >> Rxg6+ Kh8 11. h5 Kh7 >> 19 826:19 0.67 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qb4 9. Rxg6+ Kh8 >> 10. Rg4 Qc5 11. Qf7 f5 >> 19-> 849:42 0.67 1. Rxh5 c5 2. Rg1 cxd4 3. Be5 f6 4. >> Bxd4 Rxd4 5. Qxc8 Qxb3 6. Qe8 Rd1 7. >> Rh6 Rxg1+ 8. Kxg1 Qb4 9. Rxg6+ Kh8 >> 10. Rg4 Qc5 11. Qf7 f5 >> >>So Crafty likes 1...c5 too. >>Kind regards >>Bernhard > >I am wondering what changes you have done with the extentions of crafty. >I assume it is to lighten the problem that crafty often completely gets >lost in extentions and never seem to finish certain plys. At least this >happens a lot when I analyze for a long time with crafty. > >If you have some settings that help this I would very much like to try them. > >Best regards >Andreas Stabel IMHO the current setting is pretty good for playing blitz and bullet, or even standard games using a mediocre pc. However, my main concern is solving positions using Crafty as a tool. So it looks possible to get a better evaluation by computing more plies with less extensions. Of course it depends on the position. Currently I try: ext/onerep=0.6 ext/incheck=0.75 ext/recapture=0.6 ext/pushpp=1. ext/threat=0.6 ext/singular=0.6 Default is 1 for incheck and pushpp and .75 for the others. May be you will do some experiments and maybe you will get better settings for your purposes. Kind regards Bernhard
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.