Author: Robert Hyatt
Date: 06:55:20 05/07/01
Go up one level in this thread
On May 07, 2001 at 04:56:53, Vincent Diepeveen wrote: >On May 07, 2001 at 03:19:05, Uri Blass wrote: > >>On May 06, 2001 at 23:53:59, Robert Hyatt wrote: >> >>>On May 06, 2001 at 19:46:43, Vincent Diepeveen wrote: >>> >>>>On May 06, 2001 at 02:28:14, Uri Blass wrote: >>>> >>>>>I gave Deep Fritz to analyze similiar number of nodes to Deeper blue and Deep >>>>>Fritz seems to be clearly better in tactics. >>>>> >>>>>Deep Fritz needs only 191728 knodes to see the line Rf5+ Ke3 >>>>>It means only 1 second if I asuume 200,000,000 nodes per second. >>>>> >>>>>I believe that Rf5+ failed low at depth 17 for Deeper blue for the reason Ke3. >>>>>The pv of deeper blue at smaller depthes is Rf5+ Ke2 >>>> >>>>11 ply for those who are good in math and a bit more real to the world. >>> >>>Uri is correct. Unless you _still_ dispute the direct statement(s) by the >>>Deep Blue team. >>> >>> >>> >>>> >>>>>Deep Fritz probably does better extensions than Deeper blue because Deep Fritz >>>>>see big fail low at depth 16. >>>> >>>>Fritz hardly has dangerous extensions. >>>> >>>>Diep has. note i am not extending passers much. Just a bit and only >>>>now and then. >>>> >>>>The Big fail low comes at 12 ply for DIEP. Then it sees Rf5 is losing >>>>because of Ke3 though it initially wants to go e3. Then i did a state >>>>check to see what the deepest search lines are. You can see it >>>>yourself: >>> >>> >>>What does any of this matter? Their score was bad... yours is bad, black >>>is lost... I don't see where you see it any faster than they did... >> >>I see that Deeper blue score is clearly better than the score of other programs >>after search. >> >>Deeper blue said only 2.1 pawns for white after 73 seconds of search when other >>programs has no problem to see clearly better score for white. >> >>I can explain 1 pawn difference or even 1.5 pawns difference by different >>evaluation but the difference between Crafty's evaluation(4.22) and their >>evaluation(2.1 after 73 seconds) is more than 2 pawns(I mean to the evaluation >>of Rf5+) and it can be explained only by the fact that crafty could see deeper. >> >>Their score at depth 15 is only 1.63 for white so if you compare same depth then >>it is clear that Crafty did better extensions than deeper blue. >> >>If you do not like depth 15 of move 43 because of the bug that cause deeper blue >>to play Rd1 you can take depth 11(6)=17 at move 42 abd you find there a score of >>only 1.36 pawns for white. >> >>I assume that 11(6) means depth 17 with futility pruning and in this case the >>top programs of today clearly do better extensions than deeper blue. >> >>Uri > >Did you take into account that pawn=128 or something in deep blue? > >Note that if 11(6) is 17 ply they somehow prune 16 ply away. >Deep Blue was said to have next extensions: > - singular extensions with 1 ply extended > - recapture extensions 1 ply extended > - consecutive moves 1/4 ply extended (if same piece moves they extend it) Where are you getting this? They did _not_ do fractional ply extensions and this "consecutive move" extension is not in any writeup I have ever seen. > - 2 moves best (form of threat extension) 1/2 ply extended Ditto. This didn't come from them. No idea where it _did_ come from. But I had already asked about fractional extensions a long while back and got a "no". > >So for some reason they do loads of extensions show similar >lines as i do at about the same depth, from tactical reason seen. > >Sometimes they need 1 ply less to show it! Vincent: here is a problem with your statement above. I know a lot about what DB does. I know a lot about what Crafty does. Until you can beat Crafty regularly, please don't continue to point out how great your search is. If it can't beat my program consistently, even though you continually point out how bad my very simple q-search is, then how can anybody take you seriously when you continually point out how much better _your_ search is than Deep Blue's? It simply isn't credible. > >It makes no sense forward pruning 6 ply if you have hardware processors >that search 6 ply! Then it makes no sense to prune _anything_. That statement has so many holes, I really don't want to take the time to explain why. If you are saying that the hardware search can't work with a selective search, that is nonsense. > >Note that in that case we would see long mainlines which we do not >see. We only see tactical lines extended. Positional lines we can all >explain with the above extensions! I don't believe I have _ever_ seen anyone so hard-headed. One more time: The hardware will _never_ show a PV. It can _never_ do it. There is no way to get the last 6 plies + captures out of the hardware. So when you see 11(6) you are going to see 11 plies of software search + possibly some extensions, and most of the time the PV is not even going to show all of that since the positions are scattered around in the memory of 32 machines, and reconstructing the PV is non-trivial when you have to deal with that. So get over the fact that their PV will _never_ be as long as yours or mine. It is simply irrelevant. Belle's PVs were _always_ exactly 2 plies long, unless the first move was a check, in which case its PVs were 3 plies long. Do you want to go back to 1980's hardware and show me how easily you could beat Belle since it obviously couldn't search more than 3 plies, even though they _say_ 9 plies??? > >Even with huge pruning it's impossible to search 17 ply. No it isn't. I just searched to 15 plies in the above position in a few minutes. On an Alpha I could _easily_ do 17 plies in that same time limit, I only need a factor of 8-10 in speed, and I could get a factor of 30+. > >FHR is a very dubious way of searching. It's very incorrect. I can't >imagine DB ever used it. Why the rambling? They didn't do this. It is not mentioned _anywhere_ in any writeup about them. > >FHR basically says next: > > IF evaluation + thread >= beta > THEN reduce depth 1 ply; > >This can only apply for half of the search depth *at most*. > >The only official search depth example i have seen from Hsu was >examining a 'typical' 12 ply depth search! If you mean Murray you should have asked him. When talking about search, they almost _always_ used the software search depth in discussions, without thinking. Since that was the "best" part of their search (using SE, recapture, etc) and since they could see the PV from that part of the search, that is what they did their debugging on. The chess procesors could be thought of as simple endpoint evaluation functions, except that part of the evaluation was a 6-ply exhaustive search + capture quiescence search + checks in the q-search.
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.