Author: Robert Hyatt
Date: 08:23:37 04/24/01
Go up one level in this thread
On April 24, 2001 at 10:19:21, Vincent Diepeveen wrote: >On April 24, 2001 at 09:34:18, Terry McCracken wrote: > >>On April 24, 2001 at 08:20:57, Vincent Diepeveen wrote: >> >>>On April 24, 2001 at 03:47:15, Uri Blass wrote: >>> >>>>the best software that is not IBM. >>>> >>>>Suppose there is a match of 20 games at tournament time control >>>> >>>>I am interested to know how many people expect 20-0 for IBM >>>>How many people expect 19.5-.5?.... >>> >>>>If IBM expect to do better result then the average result that the public expect >>>>then they can earn something from playing a match of 20 games with Deep Blue. >>>> >>>>I believe that a part of the public who read the claim that kasparov played like >>>>an IM are not going to expect good result for IBM.> >>>>Uri >>> >>>First of all IBM would get out of book every game with -1.0 pawn >>>disadvantage (which is about the average of what Kure and Noomen >>>get in tournaments, sometimes they get out of book with mate in XXX even). >>> >>>I would expect IBM to lose with 18-2. >>> >>>Let's be realistic >>> >>> a) IBM searched 11-13 ply in 97, nowadays programs search deeper >>> b) their book is hell worse as nowadays books are >>> c) positionally it never was good, it doesn't even >>> know what a good bishop is nor when a doubled pawn is >>> good (f2,g2,g3 pattern happened twice in games against kasparov) >>> also it exchanges sometimes queens in a position where not exchanging >>> wins for IBM >>> d) hardly can use EGTBs >>> >>>So in *all* respects it is getting outgunned. Not to mention EGTBs. >>> >>>No one talked about that subject yet, but last so many plies they can't >>>use EGTBs. They only can use them the first 5 or 6 ply, that's it. >>> >>>Though this is very good compared to not using them, this means simply >>>that all exchanges towards a lost 5 men they will not detect. >>> >>>They lose with induction to everything. The level of software has increased >>>bigtime when compared to 1997. Of course the strategical problems are >>>still there and some positional problems are still there, but in >>>computer-computer games you hardly can take advantage of that. Only >>>a human versus a computer can! >>> >>>Best regards, >>>Vincent >> >>Hello Vincent, some of your data on IBM is faulty; >> >>(a)IMB's Deep Blue 1997 version could search up to 22-24 plys in the middlegame! >> No commercial program can do that at an average of 3min. per move. > >Actually some lines diep searches up to 60 ply easily. > >60 ply maximum search depth is for like a 9 ply search alrady valid. >Deep blue had a hardware limit of 32 ply so in that respect i always >search deeper as deep blue! DB had no such limit. I don't know where you got that from, but it wasn't from the DB team or anything they wrote. They _claimed_ to search to 45-50 plies along critical pathways due to singular extensions, which shoots that "32 ply limit" down in flames... > > >>(b) There opening book(s) were enormous and I'd bet larger than any programm >> on the market. They would not disclose how large, but maybe Joel Benjamin >> could help you out there:) > >Yes like my 10 million move book automatically generated was. >This book on average got out of book -2.0 or something against Kure / Noomen. However, _they_ had several GMs working on the opening book. I doubt Kure or Noomen would claim to be better... > >>(c)Positionally Deep Blue was very good and certainly did know a good bishop >> from a bad bishop! Yes, it tended to make some anti-positional moves but >> how many programs today still don't? > >Bad bishop goes wrong in a lot of programs, but many problems deep blue >showed they do not do wrong. > >> I think you're talking about the first game, and I doubt there is software >> on standard PC's that could do any better at this time. > >I am very sure all programs would do better. > >Because if you never play h6? and e6? then you definitely never get >into the position where all programs also play g5? > >> In one game Kasparov drew DB it was due to the fact he exchanged down Queens >> when keeping them on the board would have been most likely winning for him. >> However, DB was all to ready to swap down, so the computer erred as well. > >Let's not talk about kasparov's bad performance. However >amazingly despite kasparov's childish play he managed again to get 50%. > >Of course his last game was a stupid mistake. Probably his advisor: >frederic friedel, who knows shit from how chessprograms play, had analyzed >for him with fritz and fritz3 probably didn't play very aggressive in >those days and never would have played moves to open position. > >I'm not blaming Frederic at all in that, games before this >deep blue played very passive and happily exchanged queens. > >The only one to blame is Kasparov. > >> So I agree with you up to a point. Also I may agree with you to a point that >> DB did not always handle good versus bad bishops perfectly, but again this >> can still be a problem with chess programms, then and now. > >The search depth confusion from deep blue must get taken away forever. >Please analyze the logfiles. That is my point. Here is an excerpt from the log for game 1 in 1997: 9(6) #[Nd7](-17) -17 T=7 Nb8d7 pc2c4 Pe7e6 pc4d5P Pe6d5p ph2h3 Bg4f3n bg2f3B Ng8f6 10(6) #[Nd7](-15) -15 T=19 Nb8d7 pc2c4 Pe7e6 pc4d5P Pe6d5p qd1b3 Ng8f6 pd2d4 Nd7b6 bc1g5 Bf8e7 11(6) #[Nd7](-18) -18 T=62 Nb8d7 pc2c4 Pe7e6 pd2d4 Pd5c4p nb1d2 Ng8f6 nf3e5 Nd7e5n pd4e5N 12(6)<ch> 'b3' That 12(6) represents 12 plies of software search + all the normal extensions added to it, plus singular extensions added on top of that. Then another 6 plies of hardware search which included all normal extensions except for singular extensions. Beyond that they did a capture search with futility pruning. 12+6 is _not_ 13. If you don't believe the interpretation of 12(6) then simply email anybody on the deep blue team. > >>(d) EGTB's, really! Did you not know that IBM's Deep Blue in 1997 was plugged >> into databases with over 3,000,000 lines? Man, now why would it need to rely >> on only 5 man tablebases. Besides, Ken Tompson had also did his part for DB. > >In hardware you can't adress EGTBs. Certainly you can. Hsu just did not. And it isn't a limiting factor. If I do a 14 ply search in Crafty, many lines extend to 15-20 plies. Yet I don't probe beyond ply 14 to control the overhead. And I don't see anybody saying that "Crafty can't use tablebases effectively." Most have said the way I do them is better than what was done early in commercial programs... > >Do you know anyway what hardware processors are in fact? > >>They lose with induction to everything. The level of software has increased >>bigtime when compared to 1997. Of course the strategical problems are >>still there and some positional problems are still there, but in >>computer-computer games you hardly can take advantage of that. Only >>a human versus a computer can! >> >>Excuse me, They lose with the induction to everthing? You must be joking! > >No not at all. > >Even against nimzo1998 i score with nowadays diep already nearly 100% score. >In 1998 nimzo 1998 was #1 at SSDF. > >I'm very sure others do the same against it. So you beat Nimzo and by that know you can beat Deep Blue? I beat my chess partner Gower all the time, so that proves I can probably beat Kasparov, I suppose? > >>As for the rest, I agree software has made a great deal of progress since >>1997 and what you say in this paragraph I mostly agree with, on _PC's_. > >Some weak points of software has been removed. > >>However, IBM's Deep Blue was a "Supercomputer" and not a "Microcomputer"! > >But let's be clear here. The first version of deep blue, chiptest. >It got 500k nps. It searched 8 ply with 500k nps. > >A program of nowadays with 500k nps gets way deeper. Acutally even >at a quad xeon i only get 100k nps and search 11-13 ply only. Chiptest didn't search 8 plies. They did 9-10. I was there. Belle did 8-9 plies at 160K nodes per second, no null-move or anything. > >But WITH more extensions as deep blue did. As i also extend loads of things >last 6 plies where deep blue only did one extra ply near the qsearch >if i read Hsu's paper very well. Note that this is a very important >extra ply, but no 'dangerous' extensions, to quote Hsu. > >>But if you think for one moment the chess programms and and the PC's they >>run on today, even the multi-processors and their respective programms, eg; >>Deep Fritz, Junior or Shredder including EGTB's, could topple Deep Blue, >>you are dreaming in "Technocolor" my friend. > >The big PR offensive from IBM definitely has spreaded a lot of lies. > >Like in a 12 men endgame some IBM PR girls announced after game where deep >blue drew kasparov (the 2 rooks versus 2 rooks endgame, >kasparov black): "deep blue played perfect >chess here as it was in its EGTBs". That is probably reasonable. I hit the egtbs when there are 16 pieces left on the board. I have occasionally hit them with 20 pieces left. 12 pieces is _definitely reaching them a lot. > >>Deep Blue should play Kramnik, but it won't happen for many reasons. > >I would have no problems with that. > >But i give you 0% chance. > >>One, yes they could lose and I think they would. > >Only DIEP would make a chance vs Kramnik. Vincent... come back to planet earth... hello........... > >Just one thing very bad in a program, like bad bishops, and you lose >games versus a well prepared humans because of it. > >Of course DIEP would be vulnerable in opening, but that's no big problem >as most likely Kramnik is not going to show his normal openings lines >anyway as he probably will think he wins anyway (like kasparov thought >and he was nearly right). > >What would your opinion on deep blue be if kasparov had won the last game? It would _still_ be very strong. It _did_ beat Kasparov earlier in the match, and drew some very difficult endgames. > >>Two, they would _Never_ agree on Kramnik having access to Deep Blue to study >>and train with before their match, they would want to keep everthing secret >>to have a chance to win as well as set it up to play Kramnik this time instead >>of Kasparov. > >>Now I think _you_ should be realistic. > >I think you know very little from what Hsu wrote technical about Deep Blue, >you never studied its logfiles and you know nothing from computerchess in >general. > >Not to mention algorithms! > >Really those logfiles from Deep Blue show a lot! Why aren't you reading them? Instead of quoting the _wrong_ search depth? > >Best regards, >Vincent > >>Best Regards, >> Terry McCracken
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.