Author: Thorsten Czub
Date: 15:36:46 09/16/99
Go up one level in this thread
On September 16, 1999 at 13:26:07, Ratko V Tomic wrote: >>64 MB for hash both, that gives hiarcs 40MB overall, >>no autoplayer, played manually. > >H7.32 on 64Mb Windows98 shouldn't have 40Mb cache. That will end >up swapping meory pages to disk file and slow it down well below >the speed without any cache NONSENSE. It is only swapping ONCE. the first move it uses hash. you have to take back the move. and let it run again. then it will not swap again. > (depends also how much other stuff >Win98 has running simultaneously, use MSINFO32.EXE utility in >"ProgramFiles\Common Files\Microsoft Shared\MSINFO32\ directory, I am not using stupid win98. >it has info what is running, which system hooks it has and option >to clear them out). > >Try also 16Mb hash on the 64Mb Windows 98. That's a safe amount on >a 64Mb Win98 machine, it works well for H732 without swapping to disk >on a fully loaded Win98. :-)) i could use 64kbyte to see no swapping :-)) >Running a single quick test (to see nps in >a known position) may not uncover the periodic swaps due to background >activity of the system. Another test is to turn off the virtual memory >swapping and see how far you can boost the hash size before the >program crashes (it actually does that when the hash is too large >and virtual memory swapping on Win98 disabled). > >As for the CB AutoPlayer cheating, that's in principle possible, but >if it is happening against your program under your tests, it should be >possible to verify that. For example, you can set up your program to log >times and all commands it receives from the AP. You can also set it up so >it takes only the moves and no other commands (to change settings or info >calls). Also you can turn off any display of your current lines & thinking, >since such stuff can be captured and passed to the other program and make >it play better (instead of making your program play worse; you can even >make program display the worst line as the best to fool the CB program >if that's how it gets the edge). With these measures in place, you can >verify whether the CB AP is doing something odd (e.g. too many dummy calls >or something to take away your time slices or memory and force needless >disk swaps or read your search info screens). Ultimately you can always >compare nps, evaluations & response by simulating by hand moves and >timings from the suspicious games (you may have to either turn off any >randomizer you use or reseed it same way for test). > >In any case, while I can believe that CB Autoplayer can cheat, I can't >believe that it can cheat you repeatedly with your own program (to whose >code you have access), in your own tests and that you can't find out how. it is not a question of believing. it depends how much you have worked with autoplayers BEFORE and how much you work with cb-autoplayer and how many games you play without autoplayer at all.
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.