Author: Uri Blass
Date: 02:52:20 07/10/04
Go up one level in this thread
On July 09, 2004 at 20:47:34, Dann Corbit wrote: >On July 09, 2004 at 19:10:43, mike tubbs wrote: > >>This program seems to be everyone's punching bag, i wonder if it is even >>2200elo? > >It is about as strong as Yace and Thinker, in one test that I ran. Uri has had >some hardware problems. In a handful of games, anything can happen. > > Program Elo + - Games Score Av.Op. Draws > 1 Ruffian_210 : 2709 112 164 24 79.2 % 2477 25.0 % > 2 Crafty-1914 : 2626 122 159 25 68.0 % 2495 16.0 % > 3 DeepSjeng : 2614 124 163 25 66.0 % 2499 12.0 % > 4 List512 : 2614 124 163 25 66.0 % 2499 12.0 % > 5 Delfi : 2582 130 124 25 62.0 % 2497 28.0 % > 6 Aristarch 4.50 : 2575 130 124 25 62.0 % 2490 28.0 % > 7 Smarthink-017a : 2570 136 122 24 60.4 % 2497 29.2 % > 8 Thinker : 2556 139 113 24 58.3 % 2497 33.3 % > 9 Moveinow : 2522 143 126 25 54.0 % 2494 20.0 % >^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ >This should be very similar to the version he is running at WCCC > >10 Rebel12_CB : 2481 109 146 24 45.8 % 2510 33.3 % >11 Yace : 2475 109 146 24 45.8 % 2504 33.3 % >12 Ufim500 : 2439 113 139 24 41.7 % 2497 33.3 % >13 Movei 8.174 : 2437 115 133 25 40.0 % 2508 32.0 % >^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ >This is his old version that you can download, about 85 Elo below the current. > >14 Gothmog_10-b7 : 2426 102 133 25 40.0 % 2497 40.0 % >15 ELChinito 3.25 : 2425 124 130 25 38.0 % 2510 28.0 % >16 Jonny 2.62 : 2403 150 127 25 36.0 % 2503 16.0 % >17 Fruit-1_5-intel: 2374 178 122 25 32.0 % 2505 8.0 % >18 Tcb0050-win32 : 2349 147 121 24 29.2 % 2504 25.0 % >19 Patzer_361 : 2324 143 114 25 26.0 % 2506 28.0 % > >Of course, I suspect that you don't have much interest in data. I think that it is weaker than it mainly because of the following problems: 1)I believe that you tested with ponder off and it has a bug in pondering(not bug that cause it to lose on time but bug that can cause it to use a long time of more than 30 minutes for one move even in 2 hours per game. In the first game against Sos it used more than 30 minutes for the 7th move. Peter berger already reported about this bug but the good results in Leo's tournament were misleading me to believe that there is no problem in pondering and trying to finish the iteration is the right thing(I still believe it but the problem is slightly different and I will describe it later). I usually test with ponder off and I tested with ponder on only to see that there are no crashes. It seems that the problem is not very important for playing strength after seeing movei's results in Leo's tournament. The problem happens only when movei predict the correct move and changed its mind during pondering in the last iteration or failed low or failed high in the last iteration. Descreption of the problem: When Movei ponder the correct move it played immediately in case that it pondered enough time and did not failed high or fail low in the last iteration. In case of fail high or fail low it tries to finish the iteration. The problem is not with finishing the iteration but that it may decide that it needs to do another iteration. The reason for this problem is that there is varaible that is called ponder_enough that was supposed to get 1 after pondering enough time but it never gets the value 1. When it get the value 0 movei may play immediatly in case of using enough time but after fail low or fail high it is changed to -1 and later back to 0 so it does another iteration and that iteration is the last iteration because now it is in playing mode and decide that this iteration is the last iteration. I guess that better time management with ponder on can give it something like 20 elo. After all using more time may help and even if it does not help it still can use the opponent time and usually the opponent has not more than twice the time that movei has so the problem is not equivalent to being 2 times slower that is about being 50 elo weaker in long time control(having 1/2 of the time of the opponent is still better than being twice slower because the side with 1/2 of the time can use the opponent time). I simply saw that movei's time management was strange in WCCC from the first round. I cared about other problems for the second game like telling movei to believe that it really has less time then it really has and also mislocated my leg so I had not a lot of time near the computer to try to solve the problem after the second round. Now hopefully the situation of my leg is better(still not 100% ok) and hopefully I fixed the problem. Uri
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.