Author: Eran
Date: 05:15:48 02/08/02
Go up one level in this thread
On February 08, 2002 at 02:11:17, Uri Blass wrote: >On February 08, 2002 at 02:07:31, Uri Blass wrote: > >>On February 07, 2002 at 11:25:17, Eran wrote: >> >>>On February 07, 2002 at 08:30:07, Uri Blass wrote: >>> >>>>On February 06, 2002 at 19:48:13, Eran wrote: >>>> >>>>> >>>>>The interesting comparative results of the MATS test set (The MAstricht Test >>>>>Set) are shown below. >>>>> >>>>>Fritz7 with futility pruning (selected): >>>>>Result 18 out of 24 = 75% Average time = 18.05s / 163.09s >>>>> >>>>>Fritz7 without futility pruning (deselected): >>>>>Result 19 out of 24 = 79% Average time = 25.08s / 145.04s >>>>> >>>>>Fritz7 with latest update (version 7.0.0.6 - January 22, 2002) >>>>>MATS test set contains 24 positions in total >>>>> >>>>>For more information about MATS test set, please click below >>>>> >>>>>http://www.cs.rulimburg.nl/~uiterwyk/icca/mats.htm . >>>>> >>>>>Also, the test set in epd format can be downloaded from there. >>>>> >>>>>I tried to compare how well Fritz7 played with and without futility pruning. >>>>>Please see the comments as follows below. >>>>> >>>>>Firstly, based on the test Fritz7 without futility pruning seemed to solve more >>>>>positions than with futility pruning. The only game number 10 in the test set >>>>>was solved by Fritz7 without futility pruning only. The rest were the same but >>>>>with futility pruning Fritz7 searched a bit faster. That indicated that without >>>>>futility pruning Fritz7 seemed to play better in positional games. Obviously, >>>>>with futility pruning Fritz7 overlooked it because futility pruning was an >>>>>aggressive, faster and deeper search that might miss some good moves. Secondly, >>>>>Fritz7 without futility pruning seemed to search slower than with futility >>>>>pruning but overlooked less. >>>>> >>>>>Conclusion: I believe that Fritz7 with futility pruning is better in >>>>>time-control games in tournament such as rapid chess, sudden death and any >>>>>tactical-oriented games. On the other hand, if you want to let Fritz7 analyze a >>>>>position or a game for a long time, futility pruning should be turned off since >>>>>time is not very important. >>>> >>>>I suspect that you did not give Fritz enough time in order to know what happens >>>>at long time control. >>>>I think that it may be more interesting to see results of Fritz at 1 hour per >>>>move with and without futility pruning. >>>> >>>>total 48 hours for all the test >>>>average time of 20 seconds per position means that there was no position when >>>>Fritz needs more than 10 minutes to solve. >>>> >>>>I do not believe that there is no position when Fritz needs more than 10 minutes >>>>to solve and I guess that if you give Fritz one hour per move you will find more >>>>information. >>>> >>>>I tend to believe that the default option is best for all time control unless I >>>>see a proof that it is not and the results are not enough to convince me. >>>> >>>>Uri >>> >>>If the futility pruning cuts one variation that leads to finding the best move >>>1.b4 in the game # 10 in the MATS test set, then Fritz7 will never search that >>>variation again because it is cut forever. So I don't believe that if Fritz7 >>>with futility pruning continues searching, it will ever find the best move 1.b4 >>>in one or two hours later. I do not think that a very long time calculation for >>>Fritz with futility pruning makes sense. >> >> >>You are clearly wrong in your assumption >> >>For analysis of Fritz7 with futility pruning see >>http://www.icdchess.com/forums/1/message.shtml?212445 >> >>Fritz7 needs more than an hour but it finds b4 >> >>Uri > >I see that Fritz7 analyzed the position twice in 2 different computers. >In both cases it found b4 at depth 19. > >Futility pruning can help the program to be faster or slower in finding moves >but it is not selective search that prevent the program to see moves forever. > >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.