Author: Robert Hyatt
Date: 08:06:28 05/05/01
Go up one level in this thread
On May 05, 2001 at 01:29:40, Uri Blass wrote: >On May 05, 2001 at 00:53:39, Robert Hyatt wrote: > >>On May 05, 2001 at 00:20:16, Peter Kappler wrote: >> >>>On May 04, 2001 at 17:25:55, Robert Hyatt wrote: >>> >>>>On May 04, 2001 at 16:29:35, Uri Blass wrote: >>>> >>>>>On May 04, 2001 at 14:49:57, Robert Hyatt wrote: >>>>> >>>>>>On May 04, 2001 at 14:10:59, Uri Blass wrote: >>>>>> >>>>>>>On May 04, 2001 at 13:41:08, Robert Hyatt wrote: >>>>>>> >>>>>>>>On May 04, 2001 at 13:33:55, Uri Blass wrote: >>>>>>>> >>>>>>>>>On May 04, 2001 at 13:20:51, Dann Corbit wrote: >>>>>>>>> >>>>>>>>>>On May 04, 2001 at 10:52:52, Robert Hyatt wrote: >>>>>>>>>> >>>>>>>>>>>On May 03, 2001 at 21:03:58, Dann Corbit wrote: >>>>>>>>>>> >>>>>>>>>>>>On May 03, 2001 at 18:51:08, Eduard Nemeth wrote: >>>>>>>>>>>> >>>>>>>>>>>> 12 54.72 -- 1. ... Qxa3 >>>>>>>>>>>> 12 2:00 -3.04 1. ... Qxa3 2. Bf6 Rfd8 3. Bxd8 Rxd8 >>>>>>>>>>>> 4. Rd1 Rd5 5. Qe4 g6 6. Qb4 Qa6 7. >>>>>>>>>>>> Qf4 >>>>>>>>>>>> 12-> 3:19 -3.04 1. ... Qxa3 2. Bf6 Rfd8 3. Bxd8 Rxd8 >>>>>>>>>>>> 4. Rd1 Rd5 5. Qe4 g6 6. Qb4 Qa6 7. >>>>>>>>>>>> Qf4 >>>>>>>>>>>> 13 3:43 -- 1. ... Qxa3 >>>>>>>>>>>> 13 5:06 0.00 1. ... Qxa3 2. Bf6 Rfc8 3. Bxg7 Kxg7 >>>>>>>>>>>> 4. Qf6+ Kf8 5. Rf1 Rc7 6. Bg6 Re8 7. >>>>>>>>>>>> Qh8+ Ke7 8. Qf6+ Kf8 >>>>>>>>>>>> 13 11:27 ++ 1. ... b4!! >>>>>>>>>>>> 13 13:05 -0.76 1. ... b4 2. cxb4 Qd5 3. Rf1 Rae8 4. >>>>>>>>>>>> Bf2 f5 5. exf6 Qxg2 6. Bd4 Rf7 7. Bxa7 >>>>>>>>>>>> Qd5 >>>>>>>>>>>> 13 13:40 ++ 1. ... Rfe8!! >>>>>>>>>>>> 13 14:40 -2.30 1. ... Rfe8 2. Rf1 Qf8 3. Qe4 g6 4. >>>>>>>>>>>> Kb2 Rec8 5. Bf6 Rc7 6. Qf4 Rd7 7. h4 >>>>>>>>>>>> a5 8. Be4 >>>>>>>>>>>> 13 15:40 -2.63 1. ... Rfc8 2. Qe4 g6 3. Rf1 Qf8 4. >>>>>>>>>>>> Kb2 Rc7 5. Qg4 c5 6. Bxb5 Rb8 7. g3 >>>>>>>>>>>> Be4 >>>>>>>>>>>> 13-> 16:21 -2.63 1. ... Rfc8 2. Qe4 g6 3. Rf1 Qf8 4. >>>>>>>>>>>> Kb2 Rc7 5. Qg4 c5 6. Bxb5 Rb8 7. g3 >>>>>>>>>>>> Be4 >>>>>>>>>>>> time=16:39 cpu=100% mat=-3 n=596994678 fh=91% nps=597k >>>>>>>>>>>> ext-> chk=29940411 cap=1174947 pp=514823 1rep=4254557 mate=419833 >>>>>>>>>>>> predicted=0 nodes=596994678 evals=102285376 >>>>>>>>>>>> endgame tablebase-> probes done=0 successful=0 >>>>>>>>>>>>Black(1): quit >>>>>>>>>>>>execution complete. >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>Mine looks a bit different on the quad. 2:18 to drop Qxa3. Note that I used >>>>>>>>>>>hash=192M for the run... >>>>>>>>>>> >>>>>>>>>>> 12 38.66 -- 1. ... Qxa3 >>>>>>>>>>> 12 1:27 -1.65 1. ... Qxa3 2. Bf6 Rfc8 3. Bxg7 Kxg7 >>>>>>>>>>> 4. Qf6+ Kf8 5. Bg6 Rc7 6. Rf1 Re8 7. >>>>>>>>>>> Qh8+ Ke7 8. Rxf7+ Kd8 9. Qxe8+ Kxe8 >>>>>>>>>>> 10. Rxc7+ Kd8 11. Rxb7 Qxc3 12. Rxa7 >>>>>>>>>>> Qxe5 >>>>>>>>>>> 12 2:18 ++ 1. ... a5!! >>>>>>>>>>> 12 3:39 -2.57 1. ... a5 2. Bf6 Rfc8 3. Qg4 Qf8 4. >>>>>>>>>>> Qe4 g6 5. Kb2 Qc5 6. Rd1 Ra6 7. Qf4 >>>>>>>>>>> 12 4:07 -2.58 1. ... Rfb8 2. Qe4 g6 3. Rf1 Qf8 4. >>>>>>>>>>> Kb2 c5 5. Qg4 g5 6. Bf2 c4 7. Be2 >>>>>>>>>>> (4) 12-> 4:18 -2.58 1. ... Rfb8 2. Qe4 g6 3. Rf1 Qf8 4. >>>>>>>>>>> Kb2 c5 5. Qg4 g5 6. Bf2 c4 7. Be2 >>>>>>>>>>> (3) 13 5:25 -2.52 1. ... Rfb8 2. Kb2 Rc8 3. Rf1 Rc7 4. >>>>>>>>>>> Qe4 g6 5. Rxf7 Rxf7 6. Qxg6+ Kf8 7. >>>>>>>>>>> Qxh6+ Kg8 8. Qg6+ Rg7 9. Qxe6+ Rf7 >>>>>>>>>>> 13 7:40 -2.54 1. ... Rfc8 2. Rf1 Qf8 3. Qe4 g6 4. >>>>>>>>>>> Kb2 Rc7 5. Bf6 c5 6. Qg4 c4 7. Be2 >>>>>>>>>>> Rd7 8. h4 >>>>>>>>>>> 13 8:00 -2.55 1. ... Rfe8 2. Kb2 Rab8 3. Rd1 Qf8 >>>>>>>>>>> 4. Qe4 g6 5. Bf6 c5 6. Qg4 c4 7. Be2 >>>>>>>>>>> Qc5 >>>>>>>>>>> (3) 13-> 8:00 -2.55 1. ... Rfe8 2. Kb2 Rab8 3. Rd1 Qf8 >>>>>>>>>>> 4. Qe4 g6 5. Bf6 c5 6. Qg4 c4 7. Be2 >>>>>>>>>>> Qc5 >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>I will have some 64-cpu alpha numbers in a month or two. Working on a port >>>>>>>>>>>to use UPC right now... Compaq is loaning me a single-cpu alpha to compile/test >>>>>>>>>>>on with the target of a 64 cpu machine they have. I will try to get it on to >>>>>>>>>>>ICC on a weekend maybe... Or maybe for the next CCT. :) >>>>>>>>>> >>>>>>>>>>Prophecy: >>>>>>>>>>You will win the next WCCC[*] >>>>>>>>>> >>>>>>>>>>[*] Unless someone else does the same port. There is no other machine that even >>>>>>>>>>comes close. >>>>>>>>> >>>>>>>>>I am not so sure that it is enough to win. >>>>>>>>>In the last 2 WCCC tournament the biggest hardware did not win. >>>>>>>>> >>>>>>>>>Deep thought failed to win in 1995(Fritz3 was the champion) >>>>>>>>>Deep Junior,Deep Fritz,Ferret failed to win in 1999 and Shredder won. >>>>>>>>> >>>>>>>>>Uri >>>>>>>> >>>>>>>> >>>>>>>>In the last two WCCC's there were no programs that were _really_ searching >>>>>>>>at 60M nodes per second either. :) >>>>>>> >>>>>>>Yes but in the WCCC of 1995 Fritz was also clearly slower and I also believe in >>>>>>>diminishing returns so 2M against 60M is not the same as >>>>>>>0.1M against 3M. >>>>>>> >>>>>>>Uri >>>>>> >>>>>> >>>>>>I don't believe in "diminishing" returns when it is computer vs computer. Give >>>>>>me that extra ply _any_ day. It will swing the match in my favor if my opponent >>>>>>and I are equal at equal search depths. >>>>> >>>>>I believe in diminishing returns between different programs for the same reason >>>>>that diminishing return may happen in comp-human games. >>>>> >>>>>At small depthes tactics dominates so the 30 times fastesr program usually wins. >>>>> >>>>>At big depthes there are things that one program understands and the second >>>>>program does not understand when depth is not going to help. >>>>> >>>>>If 2 different programs have different positional weaknesses then the slower >>>>>program has practical chances to win at big depthes. >>>>> >>>>>Uri >>>> >>>> >>>> >>>>OK... then at _today's_ computer speeds, I don't believe in diminishing >>>>returns yet. In 20 years, perhaps. But the difference between a 15 ply >>>>search and a 17 ply search is _significant_ still. Lots of experiments have >>>>shown that diminishing returns don't appear to happen at any depth we can >>>>reach today, even using 24 hours of computer time. >>> >>> >>>What about Ernst Heinz's fixed-depth, self-play matches with Fritz? They seemed >>>to strongly suggest diminishing returns, even at depths much shallower than 15 >>>or 17 plies. >>> >>>-Peter >> >> >>Perhaps the program? Hans Berliner did an interesting experiment a long while >>back, and concluded that "dumber" programs show this diminishing return problem >>sooner than "smarter" programs. > >I do not think that Fritz6 is dumb. >A dumb program has no chance to lead the ssdf list that is no blitz. There I totally disagree. It is _clear_ that older versions of Fritz were definitely "dumb" whether the current version is or not. And _they_ were able to hit the top of the SSDF as well. To beat other computers is a different thing than to compete with human GM players... > >The result that I saw in Ernst Heinz's book do not prove diminishing return but >they suggest that diminishing returns happens and you simply do not have enough >games to prove it. It might be that when we can reach depths beyond 20 plies in a normal game, this might begin to happen. I don't know with no data to support it. But we aren't going to hit 20 plies for another two decades or so... > >I believe that the main problem is that the diminisihing return is not big >enough and you need more games to prove it. That is possible. But if it is 20+ years off before it becomes a problem, then I don't consider it an issue today... > >I agree that programs with less knowledge in the evaluation should show bigger >diminishing returns but I still believe that for all programs there is a >deminshing returns. I believe there is life on other planets too. We just don't have any proof yet and probably won't for a long time. > >I believe that part of the deminishing returns is result of opening book but not >all of them. >I decided to try the nunn2 match to find if there is diminishing returns not >because of opening book(I believe that part of the openings gives no practical >chance to win for both sides if you search deep enough and I believe that the >nunn2 position are not positions when it is a practical problem). > >I do not find diminishing returns but I did not use a lot of computer time for >it and tested only depthes that are at most 9. > > Ernst also concluded that for the time being, >>at least thru 15-16 plies, there was no apparent 'diminishing returns' for his >>program when he replicated the tests Monty and I did... > >These tests are not games and the only valid test is games. >I guess that Ernst is going to find no apparant deminishing returns also for >fritz6 by the tests that Monty and you did. > >Uri I'd bet he would. However, when you look at the data, and if you believe the premise that says a search 1 ply deeper is more accurate than the previous search, even if we don't agree by how much, then the experiment showed that programs continue to change their minds at a near constant rate for many more plies beyond today's hardware capability.
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.