Author: Robert Hyatt
Date: 06:51:04 08/30/99
Go up one level in this thread
On August 30, 1999 at 02:07:16, blass uri wrote: >On August 29, 1999 at 13:55:58, Robert Hyatt wrote: > >>On August 29, 1999 at 11:15:52, blass uri wrote: >> >>>On August 29, 1999 at 10:42:03, Robert Hyatt wrote: >>> >>>>On August 29, 1999 at 10:06:40, blass uri wrote: >>>> >>>>>On August 29, 1999 at 06:29:21, Ed Schröder wrote: >>>>> >>>>>>>Hello Ed, >>>>>> >>>>>>Hello Frank, >>>>>> >>>>>>>>We (the programmers) can argue what we want Bob but this is a lost case >>>>>>>>on before hand because the formula of playing 2 programs on one machine >>>>>>>>is too good to be true. People are not going to give this up. >>>>>>> >>>>>>>>Same story as with book-learning, it hides the real strength of a chess >>>>>>>>engine. Still people take the numbers for real. Another lost case :-) >>>>>>> >>>>>>>I play more than 2000 games on one machine, looked in the LOG-File, see the >>>>>>>games and I can not say that this games the formula 2 is. >>>>>>> >>>>>>>You and Bob say that this the formula 2 is. I mean that permanent brain is not >>>>>>>importent for matches with longer time control. Its 20-40 ELO not more. 30% >>>>>>>Ponder hints pro match, and from this 30% 3% moves that are better and 1% >>>>>>>moves >>>>>>>that are not better with permanent brain (matches with longer time controls on >>>>>>>fast PCs). >>>>>>> >>>>>>>When I play with an fast processor and the engine come under tournament >>>>>>>time to >>>>>>>13/01 this engine come with an AMD K6-3 2000 MHz to 13/05 (I think). And with >>>>>>>ponder or not with ponder I become not (in der Regel, in german) an better >>>>>>>move. >>>>>>> >>>>>>>I can not see in the WinBoard debug files problems with time control without >>>>>>>ponder. >>>>>>> >>>>>>>This is not a formula 2, this is formula 1,5 with Schumacher in position 1 >>>>>>>:-))) >>>>>>>and Hyatt and Schröder in position 21/22 ! >>>>>>> >>>>>>>But a forumula 1,5 with good statistic and results. >>>>>>> >>>>>>>This is for me suspect, suspect we your statement about more ELO by using >>>>>>>Table-Bases. I think that 4-pieces make 20-30 ELO and 5-pieces make 40-50 ElO, >>>>>>>not 5 ELO ! >>>>>>> >>>>>>>Other programmer thinking we I and other programmer thinking in the question >>>>>>>about matches on one PC we I. >>>>>>> >>>>>>>I can give all logfiles from the WT-5 tournament and you can looking. >>>>>>>That´s no >>>>>>>proof of what you have been claiming, I will see an proof and I have this >>>>>>>proof >>>>>>>when I looked my results and in the log file form the WB Engines. >>>>>>> >>>>>>>OK, better are matches with 2 PCs, but for testing and playing with 2 >>>>>>>engines is >>>>>>>one PC enough and the results are interestet and good for all people that we >>>>>>>play tournaments. >>>>>>> >>>>>>>And when make Ed Schröder an Rebel Decade WinBoard Engine for more and more >>>>>>>WinBoard Fan`s ? >>>>>>> >>>>>>>Best wishes >>>>>>>Frank >>>>>>> >>>>>>>In german for Ed ! >>>>>>>Ist mir auf englisch zu kompliziert. >>>>>>> >>>>>>>Ed, stelle Dir mal folgende Frage ! >>>>>>> >>>>>>>Wenn bei einer Engine aufgrund Permanent Brain Treffer das Zeitmanagment >>>>>>>verändert wird und es zu Zügen kommt welche schneller ausgespielt werden >>>>>>>oder zu >>>>>>>Zügen welche langsamer ausgespielt werden hebt sich das wieder auf wenn vor >>>>>>>der >>>>>>>Zeitkontrolle doch wieder eine vernüftige Restzeit zur Verfügung steht. Mit >>>>>>>anderen Worten muß die Engine sich für Züge mehr Zeit gelassen haben und hat >>>>>>>dann auch Vorteile erzielt. Vorteile und Nachteile ! >>>>>> >>>>>>> >>>>>>>Crafty blitzt nicht die letzten Züge von der Zeitkontrolle (Matches auf >>>>>>>einen PC >>>>>>>ohne Ponder) und hat z.B. bei 40 Zügen in 40 Minuten immer noch >>>>>>>durchschnittlich >>>>>>>10 Minuten für die Züge 30-40 ! >>>>>>> >>>>>>>Daher verstehe ich die Äußerungen nicht, denn es gibt ja dann auch Vorteile. >>>>>>>Vorteile weil für Züge auch eine längere Zeit zur Verfügung steht. Das muß >>>>>>>doch >>>>>>>absolut logisch sein. Ich denke nicht das dies statistisch gesehen relevant >>>>>>>ist. >>>>>>> >>>>>>>Es sind keine zwei PCs mehr notwendig für Engine-Engine Vergleiche ! >>>>>>>Für 20-40 ELO ? Diesen Nachteil haben alle Programme ! >>>>>>> >>>>>>>Gruß >>>>>>>Frank >>>>>> >>>>>>I agree all programs have this problem but you overlook one important thing >>>>>>which is my main complaint to make engine-engine on one PC being trustworthy. >>>>>> >>>>>>Due to the lack of the permanent brain the "time control" (TC) gets messed >>>>>>up. TC is an important part of a chess program. Chess programs for instance >>>>>>are keen to keep a certain amount of spare time in case the program finds >>>>>>itself in trouble (dropping score etc.). Without a permanent brain this "spare >>>>>>time" case is going to fail as the permanent brain definitely is a part of it. >>>>>> >>>>>>This is just one example. I am sure that in every program TC is done in >>>>>>different ways as there are many things involved in TC. >>>>>> >>>>>>To compete in engine-engine on one PC the program needs a *special* >>>>>>TC that takes care of the lack of the permanent brain. Next the program >>>>>>needs a piece of smart software that automatically detects that it is forced >>>>>>to play without its permanent brain because it is unlikely the user has set >>>>>>the permanent brain to "off" for the match, right? >>>>>> >>>>>>The bottom line: program_X may have all done this and program_Y not. If >>>>>>so program_X will have a very big advantage. I estimate it at 50-100 elo. >>>>>>And how can you know that if it is done or not? >>>>> >>>>>I think it is clearly less than 50-100 elo. >>>>>50-100 elo difference is the difference between p200 and p90(see ssdf results). >>>>> >>>>>If I assume that you have 1.5 minutes per move instead of 3 minutes per move for >>>>>moves 31-40 then you are 2 times slower only for 10 moves and faster for the >>>>>first 30 moves so you lose clearly less than 50-100 elo. >>>>> >>>>>I estimate the difference is 20 elo if only one program is prepared to games >>>>>without permanent brain >>>>> >>>>>Uri >>>> >>>> >>>>No... you are missing his point. playing without thinking on the opponent's >>>>time is 'unnatural' in that we don't test this way. At least most of us >>>>don't. So it is possible that program A has been tested/tuned so that it works >>>>fine without pondering, but program B might only be tested with it on. That is >>>>a big handicap to program B, and can skew the results far from what they would >>>>be on two machines... >>>> >>>>The problem is all about whether the program has been tested/tweaked to run >>>>well in that environment. It would probably be better to play on one machine >>>>using pondering instead of turning it off. And even that has problems... >>> >>> >>>I am interested to know how much do you gain in 2 computers from smart using of >>>time relative to the simple method of using the same time for every move. >>> >>>I guess that you cannot earn 50-100 elo only from smart using of time. >>> >>> >>> >>>Uri >> >> >>I would bet that if a program just uses a pure fixed time-per-move algorithm, >>it would be as much as 200 points weaker than a program that knows to (a) use >>more time when the score unexpectedly drops; (b) when the position is unstable; >>(c) when the position is more complicated, such as right out of opening book. >> >>It is easy to test, because I could take a couple of random logs and see how >>many times the score dropped by 2-3 pawns and extra time saved it. I'd bet >>it happens at _least_ once every game or two... > >The only test is games(Of course when you have x minutes per game and not per 40 >moves you cannot use the same time for every move but you can use times like >2.5% of the time per game) > >The question is if saving 2-3 pawns changed the result. >In games that you lose it clearly does not help and you cannot be sure that it >help in other games because the opponent may not see tactics and miss the >winning of 2-3 pawns and the opponent can see winning of 2-3 pawns and do a >mistake later. > >I am also interested to know if cases when the score dropped by 2-3 pawns and >extra time saved it happen more in fast time control relative to standard time >control. > >If this is the case then the difference in rating should be bigger in fast time >control. > >Uri What I am saying is that in every other game or so, I see the eval drop from (say) +.6 to -2.2, the search uses the 'extra' time it has saved up, and finds a better move that gets back to +.5 or so. Whereas without doing this, it would have had to play the -2.2 move which would probably have lost. Happens more in tactical positions or in tricky endgames.
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.