Author: Christophe Theron
Date: 11:49:09 03/30/01
Go up one level in this thread
On March 30, 2001 at 13:03:14, Robert Hyatt wrote: >On March 30, 2001 at 12:34:37, Christophe Theron wrote: > >>On March 30, 2001 at 03:29:51, Tony Werten wrote: >> >>>On March 29, 2001 at 22:38:33, Christophe Theron wrote: >>> >>>>On March 29, 2001 at 16:01:51, Tony Werten wrote: >>>> >>>>>On March 29, 2001 at 15:12:21, Christophe Theron wrote: >>>>> >>>>>>On March 29, 2001 at 14:39:34, Tony Werten wrote: >>>>>> >>>>>>>On March 29, 2001 at 14:21:16, Christophe Theron wrote: >>>>>>> >>>>>>>>On March 29, 2001 at 13:49:06, Tony Werten wrote: >>>>>>>> >>>>>>>>>On March 29, 2001 at 13:31:59, Christophe Theron wrote: >>>>>>>>> >>>>>>>>>>On March 29, 2001 at 09:14:57, Robert Hyatt wrote: >>>>>>>>>> >>>>>>>>>>>On March 29, 2001 at 06:22:13, Jouni Uski wrote: >>>>>>>>>>> >>>>>>>>>>>>On March 29, 2001 at 06:17:50, Alexander Kure wrote: >>>>>>>>>>>> >>>>>>>>>>>>>On March 29, 2001 at 04:37:19, Tony Werten wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>>Hi all, >>>>>>>>>>>>>> >>>>>>>>>>>>>>until what depth do various programs probe the tablebases ? >>>>>>>>>>>>>> >>>>>>>>>>>>>>cheers, >>>>>>>>>>>>>> >>>>>>>>>>>>>>Tony >>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>>Hi Tony, >>>>>>>>>>>>> >>>>>>>>>>>>>In London 2000, I let Nimzo 8 play with a depth of 6 plies, but later I came to >>>>>>>>>>>>>the conclusion that 8 plies might be better overall. This is indeed the default >>>>>>>>>>>>>setting of NimzoX and Varguz playing on ICC. >>>>>>>>>>>>> >>>>>>>>>>>>>Greetings >>>>>>>>>>>>>Alex >>>>>>>>>>>> >>>>>>>>>>>>Sorry one stupid question: is this the first or last 6/8 plys? >>>>>>>>>>>> >>>>>>>>>>>>Jouni >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>His statement would make no sense if it were the _last_ 6-8 plies. Those >>>>>>>>>>>are the ones that kill performance if you aren't careful. The first 6-8 plies >>>>>>>>>>>don't cost a thing. >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>But it could also mean it probes TBs in all the plies except the last 6/8. >>>>>>>>>> >>>>>>>>>>Meaning that if Nimzo is doing a X plies search, then the program probes the TBs >>>>>>>>>>in the tree for all nodes that have a distance from the root below or equal to >>>>>>>>>>X-6 (or X-8). >>>>>>>>> >>>>>>>>>Could be but that would mean, with an average depth of 10 to 12, you'd be >>>>>>>>>probing the first 4 to 6 ply. I mean, it helps but it could help more. >>>>>>>>> >>>>>>>>>That is, not counting extensions, but you if you search 10 ply, you don't know ( >>>>>>>>>at ply 8 ) how many plies are still coming. >>>>>>>>> >>>>>>>>>cheers, >>>>>>>>> >>>>>>>>>Tony >>>>>>>> >>>>>>>> >>>>>>>>But you know how many plies you have done since the root position... >>>>>>> >>>>>>>2 possibilities: >>>>>>>- We are talking about different things >>>>>>>- I don't get it >>>>>>> >>>>>>>Might be both. >>>>>>> >>>>>>>Tony >>>>>> >>>>>> >>>>>>You don't know how many extensions you are going to do, so you just ignore >>>>>>extensions and assume you are going to search full width the same number of >>>>>>plies as your iteration number. >>>>>> >>>>>>So at iteration 10, stop probing TBs after two moves from the root (assuming >>>>>>like Nimzo that you don't probe in the last 8 plies of search). >>>>>> >>>>>>That is why I say "you know how many plies you have done since the root >>>>>>position...". >>>>> >>>>>But then you don't have almost no depths where you probe. (In your example only >>>>>ply 1 and 2 ) >>>> >>>> >>>> >>>>Yes, but your NPS does not go too low then. >>>> >>>>It's a compromise between speed of search and accuracy of the evaluation. >>>> >>>>Don't forget that accessing a TB slot can be as expensive as searching 1000 >>>>nodes (depends on your program). And that many TB accesses are just a waste of >>>>time because they are not going to have any influence on your main line. >>>> >>>>I'm not saying that 8 plies like in Nimzo is the right number (actually I'm not >>>>even sure that Nimzo does it this way). >>>> >>>>In Tiger I stop accessing the tablebases a few plies before I reach the horizon. >>>>If I don't do that the program gets much weaker because of the dramatically >>>>slower NPS. >>> >>>Make your program slower ! >>> >>>XiniX is a slow searcher and I'm testing on slow hardware so I hardly notice a >>>slowdown because of probing. ( Might not be the best solution ) >>> >>>The idea I have for probing a lot is that if a position is worth being searched, >>>it's probably worth being probed. ( Unless you're very near to the leafs ) >>> >>>Hmm, search position normal, then store no nodes in hashtable, then when >>>searched second time: if n.o. nodes>1000 do probe. I'll have a try. >>>Maybe combined with: first x ply always probe. >>> >>>Positions below current position stored in hashtable can help with enhanced >>>transposition cutoffs as well. >>> >>>I'll put it in my totry book. >>> >>>Tony >> >> >>If you search 1000 positions per second and your hard disk is fast enough to >>make 1000 probes per second, don't hesitate, you can probe everywhere in the >>tree. >> >>But if you can make 100000 NPS, then probing always will weaken your program >>significantly. >> >>That's also why you need 2 days to implement TBs in a program, and 2 months to >>have them working right (improving the strength of your program). >> >> >> >> Christophe > > >Note that even though you search 1M nodes per second as I do. I _never_ probe >at that rate. Because if you only probe for captures that take you to 5 or >fewer pieces, you eliminate most of the nodes quickly and don't probe there. >The requirement that Move(ply-1)-> capture and TotalPieces(ply)<=5 _really_ >restricts the number of probes, because if you probe there and you don't get >a hit, you won't probe anywhere _below_ that node in the tree until another >capture is made... I'm also using this condition (capture && nbpiece<=5) and I guarantee I get the hard disk working real hard in endgames positions. The result on endgame test suites was a real disaster with this condition only. So I added a condition not to probe when too close (and beyond) from the horizon. Then I also added hand crafted endgame knowledge to decide in which cases a probe is useless. The result is that Tiger is hardly slower (in NPS) in the endgame, but still benefits from TBs. Christophe
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.