Author: Pete R.
Date: 13:57:32 06/07/00
Go up one level in this thread
On June 07, 2000 at 12:01:33, Pete R. wrote: >On June 07, 2000 at 01:51:18, Dann Corbit wrote: > >>On June 06, 2000 at 19:19:19, Pete R. wrote: >>>On June 06, 2000 at 16:00:17, Dann Corbit wrote: >>>>I think this is a good example to show that "best moves" in EPD test positions >>>>are only "best moves" for a certain depth unless a checkmate is found. >>> >>>I wouldn't go quite that far. For example, after 1. e4 dxe4 in this position, >>>black is completely lost (assuming best play by white of course >> >>See Bernard B.'s post... Crafty seems to have found a loophole in 1. e4 dxe4 >>... >>;-) > >I assume you mean this: > >16 356:58 -0.15 1. e4 dxe4 2. g4 Bc8 3. Qxf7+ Kxh6 > 4. Rh3+ Bh4 5. Rxh4+ Qxh4 6. Qxe8 Kg7 > 7. Qe5+ Kf8 8. Qf4+ Kg8 9. Bxe4 Rb8 > 10. Bd5+ > >Crafty is simply not doing well here (I tried 16.15 and 17.10). The winning line >is 2...Bc8 3. Qxf7+ Kxh6 4. Rh3+ Bh4 5. Qh5+ Kg7 6. Rxh4 Qf6 > 7. Qxe8 Qxh4 8. Re1 +-. > >From Bc8, it doesn't see 5. Qh5+ is best. As you go down the winning PV it does >choose Qh5+, but then from there it doesn't see 7. Qxe8. In fact if you feed it >the position at 6...Qf6, at 12 ply is still chooses Qh7+ with an eval of 0.52. >Then you feed it 7. Qxe8 and of course the capture Qxh4 is forced, and then it >quickly sees that Re1 wins for white. But missing Qxe8 from the immediately >preceeding position is very strange because it takes less than 12 ply to see >that white wins once you give it the move. Not to pick on Crafty, other programs seem to exhibit a similar effect, not finding Qxe8, even though they have searched more ply than is required for them to see that it wins once you provide it. I don't understand this sort of thing. In any case it takes a 12 ply search from 6...Qf6 to select Qxe8, so it would take an extremely deep search to find this win from move 2. It seems that once you get to a certain depth, it takes so many hours to get to the next ply that it's just a waste. Much better would be to invoke some sort of "correspondence" analysis mode where you select candidate moves at each ply and then sequentially investigate the variations as a human would. This sort of meta-searching has been discussed before, I suppose it's easier said than done. But at least for doing very deep searching that type of mode might be superior.
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.