Author: Robert Hyatt
Date: 07:52:33 07/01/98
Go up one level in this thread
On July 01, 1998 at 01:51:34, blass uri wrote: >On June 30, 1998 at 21:22:07, Robert Hyatt wrote: > >>On June 30, 1998 at 19:53:08, blass uri wrote: >> >>>On June 30, 1998 at 17:38:40, Robert Hyatt wrote: >>> >>>>On June 30, 1998 at 16:33:45, blass uri wrote: >>>> >>>>>On June 30, 1998 at 13:33:19, Robert Hyatt wrote: >>>>> >>>>>>On June 29, 1998 at 18:45:15, Vincent Diepeveen wrote: >>>>>> >>>>>>> >>>>>>>On June 28, 1998 at 20:24:15, jonathan Baxter wrote: >>>>>>> >>>>>>>>On June 28, 1998 at 20:06:52, Vincent Diepeveen wrote: >>>>>>>> >>>>>>>>>So when i said: 18-20 ply is easy to do, then people laughed at me. >>>>>>>>> >>>>>>>>>Right now, diep gets after a day of search already 18-20. >>>>>>>>>It needs around 10k * 3600 * 24 = 840M nodes. >>>>>>>>> >>>>>>>>>That's with R=3 (for most programs R=2 and R=3 make no diff, but in >>>>>>>>>Diep it does), but nevertheless, this was considered *undoable* 2.5 years ago. >>>>>>>>> >>>>>>>>>Note that this is just with 60MB for hash, and at those slow levels >>>>>>>>>a doubling of hash give another ply because of the huge load factor. >>>>>>>>> >>>>>>>>>How opinions change. So 20+ ply for Diep is easily doable with 200M >>>>>>>>>nodes a second. In fact with say 1 gig for hashtables instead of the >>>>>>>>>60M i'm using now, i'll get 20 within few >>>>>>>>>seconds. >>>>>>>> >>>>>>>>Just to see how dumb this statement is: with enough forward pruning *any* >>>>>>>>program can search the full game tree and announce draw or win or whatever on >>>>>>>>the first move, for a stated "ply" of 200+. But unless it is always right, you >>>>>>>>can't claim it is searching to that depth. End of story. >>>>>>>> >>>>>>>>Jonathan Baxter. >>>>>>> >>>>>>>Forward pruning works ok, i compensate with extensions a *lot*. >>>>>>> >>>>>>>So if i'm getting 20 ply after 24 hours, then it really is 20 ply in normal >>>>>>>positions. It just prunes out some lines where you give directly a queen >>>>>>>away. I think that the chance that giving a queen away or something like >>>>>>>that wins the game is very little. No doubt that it'll look it'll search that >>>>>>>line at least 20-4 = 16 ply. >>>>>>> >>>>>>>Still. If i turn OFF forward pruning then i get this 20 ply easily with a huge >>>>>>>hashtable. Yet i don't HAVE such a huge hashtable, nor a PII-300. >>>>>>> >>>>>>>I have at maximum 60MB hash, which gives already a *huge* depth, >>>>>>>only some crap gets deleted last few plies. >>>>>>> >>>>>>>So i'm talking about a program that practical *gets* 20 ply, just >>>>>>>a very few positions where there is a 20 ply trick at least with giving away >>>>>>>material and stuff, *might* get missed. >>>>>>> >>>>>>>Note that my mating extensions do a rather good job. >>>>>>> >>>>>>>For example win at chess 141: >>>>>>> >>>>>>>4r1k1/p1qr1p2/2pb1Bp1/1p5p/3P1n1R/1B3P2/PP3PK1/2Q4R w Qc1xNf4 WinAtChess.141 >>>>>>> >>>>>>>00:00 113 1->1 -3.87 Rh4xf4 Bd6xf4 Qc1-c5 >>>>>>>00:00 517 1->1 -2.42 Kg2-f1 Re8-e2 >>>>>>>00:00 744 1->2 -2.39 Kg2-f1 Re8-e2 Bf6-e5 >>>>>>>00:00 3266 1->3 -2.42 Kg2-f1 Re8-e2 Kf1-g1 >>>>>>>00:01 14255 1->4 -2.48 Kg2-f1 Re8-e2 Rh1-h2 a7-a5 Kf1-g1 >>>>>>>00:03 41872 1->4 2.98 Qc1xf4 Qc7-d8 Bf6-e5 >>>>>>>00:04 53482 1->5 8.56 Qc1xf4 Qc7-d8 Rh4xh5 Qd8xf6 Qf4xf6 g6xh5 >>>>>>>00:16 202646 2->6 MATE06 Qc1xf4 Bd6xf4 Rh4xh5 g6xh5 Rh1xh5 Bf4-h6 Rh5xh6 Qc7-g3 >>>>>>>Kg2xg3 Re8-e2 Rh6-h8 >>>>>>>00:20 239812 3->7 MATE06 Qc1xf4 Bd6xf4 Rh4xh5 g6xh5 Rh1xh5 Bf4-h6 Rh5xh6 Qc7-g3 >>>>>>>Kg2xg3 Re8-e2 Rh6-h8 >>>>>>>quit >>>>>>> >>>>>>>Nolot #2 >>>>>>>r4rk1/pp1n1p1p/1nqP2p1/2b1P1B1/4NQ2/1B3P2/PP2K2P/2R5 w - - bm >>>>>>>Rxc5! >>>>>>> >>>>>>>00:00 188 1->1 -3.07 Ne4xc5 Nd7xc5 >>>>>>>00:00 210 1->1 -3.06 h2-h4 >>>>>>>00:00 211 1->1 -3.00 h2-h3 >>>>>>>00:00 226 1->1 -2.85 Ne4-f6 Nd7xf6 Bg5xf6 >>>>>>>00:00 231 1->1 -2.82 Bg5-e7 >>>>>>>00:00 234 1->1 -2.68 Bg5-h6 >>>>>>>00:00 1384 1->2 -3.35 Bg5-h6 Qc6-b5 Ke2-d1 >>>>>>>00:00 1552 1->2 -2.85 Ne4-f6 Nd7xf6 Bg5xf6 >>>>>>>00:00 1855 1->3 -2.85 Ne4-f6 Nd7xf6 Bg5xf6 >>>>>>>00:00 3129 1->3 -1.50 Ne4xc5 Nd7xc5 Qf4-e3 >>>>>>>00:01 6845 1->4 -2.74 Ne4xc5 Nd7xc5 Bg5-f6 >>>>>>>00:02 23212 1->5 -3.08 Ne4xc5 Nd7xc5 Qf4-b4 Nb6-d7 Qb4-f4 >>>>>>>00:04 48858 2->6 -3.34 Ne4xc5 Nd7xc5 Qf4-b4 Nb6-d7 Bg5-f6 Nc5-d3 Rc1xc6 >>>>>>>00:09 116252 2->6 -2.80 Bg5-h6 Qc6-b5 Ke2-d1 Qb5-f1 Kd1-c2 Qf1-e2 Kc2-b1 Bc5-d4 >>>>>>>Ne4-f6 Nd7xf6 Qf4xd4 >>>>>>>00:20 244652 3->7 -2.80 Bg5-h6 Qc6-b5 Ke2-d1 Qb5-f1 Kd1-c2 Qf1-e2 Kc2-b1 Bc5-d4 >>>>>>>Ne4-f6 Nd7xf6 Qf4xd4 >>>>>>>00:33 399303 3->7 -2.27 a2-a4 h7-h6 Bg5xh6 Nb6xa4 Ne4xc5 Qc6-b5 >>>>>>>00:43 515515 4->8 -3.99 a2-a4 Nb6xa4 Ne4-f6 Nd7xf6 Bb3xa4 b7-b5 Qf4-c4 b5xa4 Qc4 >>>>>>>xc5 >>>>>>>02:16 1655957 4->8 -2.76 Bg5-h6 Qc6-b5 Ke2-d2 Bc5-b4 Kd2-e3 Qb5xe5 Qf4xe5 Nd7xe5 >>>>>>> a2-a3 Bb4-a5 Ne4-f6 >>>>>>>04:01 2836391 5->9 -3.37 Bg5-h6 Qc6-b5 Ke2-d2 Bc5-d4 Ne4-f6 Nd7xf6 Qf4xd4 Rf8-e8 >>>>>>> e5-e6 Nb6-d5 >>>>>>>06:42 4216433 5->9 -1.15 Rc1xc5 Nd7xc5 Ne4-f6 Kg8-h8 Qf4-h4 Qc6-b5 Ke2-f2 Nc5-d3 >>>>>>> Kf2-g2 Nd3-e1 Qh4xe1 Nb6-d7 Qe1-h4 h7-h5 Nf6xd7 Qb5xd7 >>>>>>>08:37 5197732 6->10 0.00 Rc1xc5 Nd7xc5 Ne4-f6 Kg8-h8 Qf4-h4 Qc6-b5 Ke2-e3 Qb5-d3 >>>>>>> Ke3-f2 h7-h5 Nf6xh5 Nc5-e4 f3xe4 Qd3-d4 Kf2-f1 Qd4-d3 Kf1-f2 >>>>>>>13:14 7548617 7->11 0.10 Rc1xc5 Nd7xc5 Ne4-f6 Kg8-h8 Qf4-h4 Qc6-b5 Ke2-e3 Qb5-d3 >>>>>>> Ke3-f2 h7-h5 Nf6xh5 Nc5-e4 f3xe4 Qd3-d4 Kf2-g2 Qd4xb2 Kg2-h3 Qb2-c3 Nh5-g3 Kh8- >>>>>>>g8 Bg5-f6 Qc3-c8 Kh3-g2 >>>>>>>33:30 18176564 8->12 2.65 Rc1xc5 Qc6xe4 f3xe4 Nd7xc5 Bg5-h6 Nc5xb3 a2xb3 Nb6-d7 >>>>>>>Bh6xf8 Ra8xf8 Ke2-e3 >>>>>>>59:01 31434222 9->13 2.47 Rc1xc5 Qc6xe4 f3xe4 Nd7xc5 Bg5-h6 Nc5xb3 a2xb3 Nb6-d7 >>>>>>>Bh6xf8 Ra8xf8 Qf4-g5 Kg8-g7 e5-e6 f7xe6 Qg5-e7 >>>>>>> >>>>>>>Nolot #10 >>>>>>>r1b2rk1/1p1nbppp/pq1p4/3B4/P2NP3/2N1p3/1PP3PP/R2Q1R1K w - - bm Rxf7 >>>>>>> >>>>>>>00:00 7 1->1 -4.06 Bd5xf7 Rf8xf7 Rf1xf7 >>>>>>>00:00 11 1->1 -4.05 Bd5xb7 Bc8xb7 >>>>>>>00:00 35 1->1 -3.94 Rf1xf7 Rf8xf7 >>>>>>>00:00 52 1->1 0.06 Nd4-f5 >>>>>>>00:00 176 1->2 -0.32 Nd4-f5 Be7-f6 >>>>>>>00:00 575 1->3 -0.34 Nd4-f5 Be7-f6 Ra1-b1 >>>>>>>00:00 1807 1->4 -0.60 Nd4-f5 Be7-f6 Ra1-b1 Nd7-e5 >>>>>>>00:01 10307 1->5 -0.72 Nd4-f5 Be7-f6 Ra1-a2 Kg8-h8 >>>>>>>00:02 24729 1->5 -0.57 Qd1-d3 Be7-f6 Nc3-e2 Nd7-c5 Qd3xe3 Qb6xb2 >>>>>>>00:03 32869 2->6 -0.58 Qd1-d3 Be7-f6 Nc3-e2 Bf6-g5 Nd4-b3 >>>>>>>00:07 83015 2->6 -0.33 Bd5-b3 Be7-f6 Nc3-d5 Qb6-c5 c2-c3 Bf6-g5 >>>>>>>00:10 119222 3->7 -0.17 Bd5-b3 Qb6-c5 Nc3-d5 Be7-g5 a4-a5 >>>>>>>00:27 342429 4->8 -0.13 Bd5-b3 Be7-f6 Nc3-d5 Qb6-c5 c2-c3 Bf6-g5 a4-a5 >>>>>>>01:08 855177 5->9 0.04 Bd5-b3 Be7-g5 Nc3-d5 Qb6-a5 Qd1-h5 Qa5-d8 Nd4-f5 g7-g6 Nf >>>>>>>5xd6 >>>>>>>02:40 2043189 6->10 0.12 Bd5-b3 Be7-f6 Nc3-d5 Qb6-a7 c2-c3 Bf6-g5 Qd1-h5 h7-h6 a >>>>>>>4-a5 >>>>>>>08:18 5325215 7->11 -0.01 Bd5-b3 Be7-g5 Nc3-d5 Qb6-d8 Nd4-f5 Nd7-c5 Nf5xe3 Bg5xe >>>>>>>3 Nd5xe3 Bc8-e6 Bb3xe6 >>>>>>>22:41 13898604 7->11 0.23 Rf1xf7 Rf8xf7 Bd5xf7 Kg8xf7 Qd1-h5 Kf7-g8 Qh5-e8 Be7-f >>>>>>>8 Nc3-d5 Qb6xd4 Nd5-e7 Kg8-h8 Ra1-f1 Qd4-f6 Rf1xf6 g7xf6 Ne7xc8 >>>>>>> >>>>>>>Try that at other programs. Diep sees Rxf7 at 11 ply which is so called >>>>>>>selective and I did NOT make any adjustments to find Rxf7. >>>>>>> >>>>>>>Some versions even found it earlier than this 11 ply, and >>>>>>>i bet that Deep Blue doesn't get it at 11 ply, although it needs >>>>>>>special processors and SP hardware in order to get 11 ply. >>>>>>> >>>>>>>Vincent >>>>>> >>>>>> >>>>>> >>>>>>Since you posted this, I thought a quick response would be in order. Bruce >>>>>>has the original data, but deep thought found this move in 2 minutes. It is >>>>>>over 100 times slower than deep blue, which puts it at about 1 second to find >>>>>>this. Quite a lousy program wouldn't you say? Deep thought also got it at >>>>>>a much shallower depth than 11 ply. Let's let Bruce fill in the blank as I >>>>>>didn't save the original discussion. But once again, you are off not just by >>>>>>a little, but by orders of magnitude... which is getting pretty typical when >>>>>>you use the term "deep blue" without having a clue about how the machine >>>>>>functions or what it does. >>>>>Junior4.6 and fritz5 could not solve the last position in a reasonable time. >>>>>I gave Junior4.6 2 hours on my pentium200MMX and it could not find Rxf7 >>>>>I gave Fritz5 30 minutes and it could not find the right move. >>>>> >>>>>I do not understand how deepthought could find the move in 2 minutes >>>>>when deeperblue could not find the draw in the position of the second game >>>>>kasparov resigned. >>>>>I do not think it is more difficult to see the draw in the position kasparov >>>>>resigned than to see Rxf7. >>>>> >>>>>Uri >>>> >>>> >>>>it's *easy*. The forced draw is *60* plies from the root position. Not 10, >>>>not 30, but a full 60 plies deep. I believe that Ed had the analysis on his >>>>home page for a while. So *no* one sees that draw, because it is too deep. >>> >>>I remember it is more than 40 plies but not 60 plies. >>>a good program can see the draw by doing enough check extensions. >>>I think that deeper blue designers did not do enough. >> >>before you can say that, you need to find a program that can demonstrate >>the draw. To date, *none* have. So that position is beyond us all. I don't >>remember the analysis precisely, but there was a 60 ply forced drawing path >>before one side *had* to repeat the position. Ed had it on his web site at >>one time, may still have. > >can diep demonstrate the draw? >> >>>> >>>>A human analyzes that draw a little differently, recognizing that the king can't >>>>get out of the box and gets checked forever... but a computer doesn't realize >>>>that until it actually does it. >>>> >>>>However, as far as deep thought finding it in 2 minutes, remember, I have tried >>>>to explain here many times just how fast and deep they search. But everyone >>>>listens to the bogus assumptions made by people like Vincent, and draw the wrong >>>>conclusions. DT and DB are not just strong, they are "strong as hell". This is >>>>but one example. >>>I agree about DB but I know Deep thought is not better than international >>>masters. >> >> >>I completely disagree. Deep Thought maintained a 2550 USCF rating over 24 games >>to claim stage 2 of the fredkin prize. It was clearly better than most IM's and >>as good as the top few (which are as good as the bottom few GM players of >>course.) >> >>Check out "scratchy" on ICC. It has the best win/lose ratio of any player >>there, and that was deep thought, not deep blue or deep blue junior. It was >>a *fierce* chess engine that crushed everyone (computer-wise) that it came up >>against, when you check the ACM tournament results starting in 1988. > >I was based on a tournament Deep Thought played in 1991 at Vienna >Deep thought achieved 2.5 points out of 7 in this touenament >with performance of 2410 >before this tournament Deep thought had better results but it was because >humans did not know the computer's weakness. >Deep thought lost one of its game in this tournament in 22 moves >(against Tiscbierek). this was a known problem. A GM helped with their book, and included this line thinking it was good. In fact, right out of book the opponent found a crushing move before deep thought had a chance. It happens. I think that they test opening lines better than that now, but they stepped into a bad problem in that game for sure...
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.