Author: Stuart Cracraft
Date: 16:14:30 09/05/04
Go up one level in this thread
On September 05, 2004 at 18:30:30, Bas Hamstra wrote: >On September 05, 2004 at 17:02:49, Uri Blass wrote: > >>On September 05, 2004 at 16:37:32, Stuart Cracraft wrote: >> >>>On September 05, 2004 at 15:58:39, Bas Hamstra wrote: >>> >>>>On September 05, 2004 at 14:56:14, Uri Blass wrote: >>>> >>>>>On September 05, 2004 at 14:30:01, Bas Hamstra wrote: >>>>> >>>>>>>My program has this in relation to checks, all conditionally compilable: >>>>>>> >>>>>>> 1) all checks-evasions in quiescence, unlimited occurrences >>>>>>> 2) all checks-evasions in main search, unlimited occurrences >>>>>>> 3) all checking-moves in quiescence, at the first ply of quiescence >>>>>>> >>>>>>>These are the three extensions. Only #1 and #2 have proved useful. >>>>>> >>>>>>Impossible. When I did 3) it solved wac141 within one second, using a mate >>>>>>extension of a full ply. It takes 30.000 nodes and 6 plies to anounce M6. >>>>> >>>>>With mate extension of a full ply your search will explode because >>>>>there are positions when there are a lot of checks and every reply to check is >>>>>a mate threat. >>>> >>>>Not in my search. But I have taken actions to prevent "stapling" of useless >>>>matethreats. I have done a full ply mate-extend in Tao 5.4 without problems. >>>> >>>>>6 plies to see the mate and only 30000 nodes. >>>>>It seems to me hard to believe it espacially when I expect your search with >>>>>full ply extension for mate to explode. >>>> >>>>See above for the explosion. I am not saying a full ply mate-ext is best, but I >>>>have used it for years and it worked fine. This is probably also a matter of >>>>what your qsearch can do. Mine can solve simple multi-move mates. >>>> >>>>>Of course checks in the qsearch helps but I still need 8 plies without mate >>>>>extensions only to find the move without mate score(it is after I decided to >>>>>reduce my evaluation based pruning). >>>> >>>>Yes, I checked and have the same result without mate-ext. >>>> >>>>>I will check with enabling mate extensions only in the first 11 plies(no need >>>>>to extend after 11 plies to find mate in 6) and check results again. >>>>> >>>>>>Without it it has a hard time and it takes very long. The point is: if your >>>>>>qsearch cannot find a simple mate in one (a capture only qsearch wont) chances >>>>>>are you will find this mate a full 3 plies later than necessary. >>>>>> >>>>>>If you implemented 3) your program should mate-extend like crazy here... >>>>>> >>>>>>Bas. >>>>> >>>>>If he implements 3 then his program will crash because it will never get to the >>>>>second ply of the qsearch because his qsearch call the main search and not the >>>>>qsearch. >>>> >>>>You are right here. When you do checks on the first q-ply and call the mean >>>>search for check-evasions, you will have unlimited checking sequences and that's >>>>bad. >>>> >>>>Bas. >>> >>>Repetition detection would take care of that. >>> >>>My search terminates and never goes anywhere near the 99 maximum ply >>>setting. Usually at 8, 9, 10, 11 ply it is no more than 4x the nominal >>>main search depth, e.g. 32, 36, 40, 41 -- I don't limit quiescence to >>>that, it just often turns out that way. >>> >>>No unlimited checking sequences noted due to repetition detection and >>>draw return in both mainsearch and quiescence. >>> >>>On the other hand, I am anxious to try a quiescence that does not >>>handoff to main search to get out of check in quiescence. I was hoping >>>to do that this afternoon. >>> >>>Stuart >> >>I guess that you did not search the right positions or did not search deep >>enough to get the 99 plies >> >>there are positions when there can be easily more than 99 plies of checks with >>no repetition. >> >>Uri > > >I have the same experience. > >Bas. Most programs I've heard of have a maximum ply setting beyond which the search can't go. How about you Uri and you Bas? Stuart
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.