Author: Howard Exner
Date: 09:10:55 11/17/98
Go up one level in this thread
On November 17, 1998 at 01:01:25, Dave Gomboc wrote: >On November 16, 1998 at 23:46:36, Howard Exner wrote: > >>On November 16, 1998 at 23:05:49, Howard Exner wrote: >> >>>On November 16, 1998 at 03:23:54, Harald Faber wrote: >>> >>>>Hi Ed, >>>>I am not sure to reach you via e-mail so I write it here too: >>>> >>>>I have seen strange behaviour in Rebel10. I have AMD-K6-200 and Laptop166MMX. In >>>>the same position the AMD has a diiferent best move with a different value than >>>>the 166MMX AT SAME SEARCH DEPTH. How is this possible, any explainations? >>> >>>I'm noticing something similar when even using the same machine with the >>>exact same settings. I tried this with the following position from the >>>BS2830 test suite. >>> >>>r4r1k/pbnq1ppp/np3b2/3p1N2/5B2/2N3PB/PP3P1P/R2QR1K1 w - - id BS2830.EPD; bm Ne4; >>> >>>The settings for Rebel 10 on my K6-233 were AntiGm=Smart, Combination=Off, >>>Selection=N and PlayStyle=N. The method I used was to play the position >>>in infinite mode, force the move after about 60 seconds, record the PV >>>results, takeback the move and then go through the process from the start. >>> >>>For 12 takebacks Rebel 10 played the moves in this order: >>>ST,A,A,ST,A,A,ST,A,ST,ST,A,ST. (ST means that Rebel 10 played the move >>>by setting itself to think in AntiGM=Strong, while A means REbel 10 chose for >>>itself AntiGM=Active). Note that the entire time I am using Anti=GM set to Smart >>>for the entire test procedure. >>> >>>This is definitely not making Rebel 10 deterministic when AntiGM is set to >>>Smart. Does this mean that when AntiGM is set to smart that different test >>>suites will have varying solution times based on what the Smart setting decides >>>among the three Choices of Strong, Active or Smart? >> >>correction here It should read ... three choices of Strong, Active or Off? >>> >>>My question is why is Rebel 10 chosing different AntiGM settings for completely >>>identical positions with the exact same settings? Should Smart mean the >>>ability to somehow assess the position and then decide on what AntiGM path >>>to follow? >>> >>>For my own testing and playing against Rebel 10 I'm using AntiGM=Strong, >>>Combination=Off, Selection=N (except for some middlegame testing when at >>>times I use Selection = 4), and PlayStyle = Agressive. > >I suspect that Ed is doing some funky stuff with whatever he's got stored in the >hash table from previous searches. Based on your ideas here I decided to investigate possible connections to the hash settings. I set hash = 0 and ran the same position over as described above. This time it plays the same way always as Smart chooses the AntiGM set to Off. Ed mentioned in a previous post, somewhat related to this, that AntiGm is a "mystery". I must agree. Another try is to run a test suite or an entire group of positions from start to finish, noting the evals. Then repeat the procedure after a reboot as you suggested. And for kicks arrange the test suite positions in a different order, then compare those evals. Will they be different or the same? >Rebel might also do position learning to >disk, so that it is persisted between sessions, I don't know. Anyway, when the >position is re-searched, the differences in whatever has been stored are causing >a different degree of anti-GM mode to be made. > >One interesting thing you might do is to power down your machine between each >trial. Another interesting thing you might do is to erase any learn files and >power down between each trial. You could then figure out a little bit about >what's happening. It's even possible that there is nothing being stored at all, >and the differences are due to a pseduorandom number generator. > >In case it's not already clear, I am merely speculating. I don't own Rebel 10 >myself. I would buy it, but I have Rebel 8 and 9 (courtesy Ed, I did some >beta-testing) and I don't use them lately because I haven't had a dos/win9x >partition in a while. As soon as the Adaptec 2940-U2W support in Linux is >stable, I will be installing Linux on my system. Does anyone know if Rebel runs >under DOSEMU?
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.