Author: Bob Durrett
Date: 10:53:51 11/18/00
Go up one level in this thread
On November 18, 2000 at 13:18:14, Uri Blass wrote: >On November 18, 2000 at 12:43:59, Stefan Meyer-Kahlen wrote: > >>On November 17, 2000 at 18:54:37, Uri Blass wrote: >> >>>On November 17, 2000 at 18:40:52, Torstein Hall wrote: >>> >>>>On November 17, 2000 at 18:33:54, Uri Blass wrote: >>>> >>>>>On November 17, 2000 at 18:23:30, Torstein Hall wrote: >>>>> >>>>>>On November 17, 2000 at 17:44:50, Uri Blass wrote: >>>>>> >>>>>>>On November 17, 2000 at 17:20:25, Fernando Villegas wrote: >>>>>>><snipped> >>>>>>>>c) If it is not, how this entity compares with Schredder IF two top programs are >>>>>>>>harnessed toguether? I suppose many experiment has been already performed before >>>>>>>>delivery. >>>>>>> >>>>>>>I think that it is has some rules based on evaluations and main lines. >>>>>>> >>>>>>>Example 1:If engine A fail low after 1 second and engine B fail low after 10 >>>>>>>seconds with the same main line then it is logical to assume that engine B is >>>>>>>weaker in tactics in the relevant position so it is logical to choose engine A. >>>>>>> >>>>>>>Example 2:If engine A shows evaluations:+1.1 and the evaluation goes down to 0.5 >>>>>>>when engine B has stable evaluation of 0.0 then it is logical to assume that >>>>>>>engine B understands the position better(I usually expect +1.1 to go up and not >>>>>>>to go down) so it is logical to choose engine B. >>>>>>> >>>>>>>I do not know if Shredder is using similiar ideas but this is the ideas that >>>>>>>seem to me logical to try in order to be correct most of the time in choosing >>>>>>>the right engine in cases when both engines have equal strength. >> >>Yes, these are some examples how it is working in principle, but this is only >>one part of the job. I also check the pvs for example. >> >> >>>>>>> >>>>>>>Uri >>>>>> >>>>>>I feel that such a beast must be weaker than a normal program, as it have to >>>>>>waste a lot of computing time. First two engines computing the same pos. Then >>>>>>the third engine etc. etc. >> >>Yes, this is the problem on one single cpu machine, both engines will only get >>50% of the cpu, but the thing gets boosted if you are running on a dual machine >>or in a network. >> >>>>>> >>>>>>Torstein >>>>> >>>>>I disagree because of the following reasons: >>>>> >>>>>1)The third engine may have a simple rule to decide so it practically may waste >>>>>less than 1% of the time. >>>> >>>>Probably true! >> >>Yes, the triple brain hardly needs any cpu time in comparison to the engines. >> >>>> >>>>>2)The two engines do not use the same time so it can be clearly less than twice >>>>>slower because the playing engine may be used 90% of the time. >>>>> >>>>This I think (or feel)must be wrong. >>>>Both engine A and B have work on the position at first for the engine 3 to >>>>deside on them. So in effect you lose a lot before you deside. And if you only >>>>switch engine rarly, I feel that somehow the idea loses much of its point. >>>>But perhaps Stefan can tell us more? >>> >>>1)Stefan explained in a previous post that the engines are not used for equal >>>time. >> >>You go me wrong. Both engines are running 100% of the time. The triple brain >>decides when to stop them simultaniously. >> >>The timing is very important. It is for example more likely that a move is easy >>or hard if both engines say so. > >Do you have cases when the tripple brain of two engines got better result than >everyone of them in comp-comp games. > >If you have cases like it then it will be interesting if you post the relevant >games(I guess that shredder5 is not one of them because it is too strong >relative to the other engines so I see no reason to hide details about it) > >Uri What would be REALLY interesting to see would be the choices made by the voting software, i.e. which moves were chosen by each engine and which were selected by the voter. Probably not available information, but would be revealing.
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.