Author: Robert Hyatt
Date: 13:41:46 05/04/04
Go up one level in this thread
On May 04, 2004 at 07:44:22, Uri Blass wrote: >On May 04, 2004 at 06:52:00, Mridul Muralidharan wrote: > >>On May 04, 2004 at 06:24:28, Uri Blass wrote: >> >>>On May 04, 2004 at 06:06:34, Mridul Muralidharan wrote: >>> >>>>On May 03, 2004 at 15:07:50, Uri Blass wrote: >>>> >>>>>On May 03, 2004 at 14:18:33, Vincent Diepeveen wrote: >>>>> >>>>>>On May 03, 2004 at 13:35:38, Uri Blass wrote: >>>>>> >>>>>>I know you do not believe a thing, especially not things where 10 people are a >>>>>>witness from in the dutch computer chess championship. >>>>> >>>>>I do not believe that these 10 people know what he does. >>>>>I do not know what he says but in case that he says that he is working a full >>>>>year only to get 0.5% speed imporovement or even 10% speed improvement(I did not >>>>>hear it from other people except you) then my assumption is that he gives >>>>>disinformation because it does not make sense to do it. >>>>> >>>>>Uri >>>> >>>>I am not sure of that Uri , I do remember Johan saying that it is getting >>>>increasingly tough to improve King. >>>>Similarly maybe for fritz - when your engine is so well optimised with author >>>>having worked on finetuning it for decades (like fritz) , then the number of >>>>possible avenues for improving it , while keeping existing strength intact, will >>>>become very tough. >>>> >>>>So maybe Frans Morch , SMK , Johan , Christoher , etc do find it tough to >>>>improve things which should have almost no negetive impact on current >>>>performance. >>>> >>>>Mridul >>> >>>SMK clearly gets more than 10 elo in a year. >>>Same for Frans morch. >>> >>>I believe that there is a big room for search improvments in chess programs and >>>programs search too many illogical lines so it is a big mistake to work on speed >>>improvement when it is possible to get more improvement by other means. >>> >>>Uri >> >>Yes , even I believe that there could still be possibly lot of improvements - >>like in better move ordering , more restricted extensions , better qsearch , >>better eval , pruning expiriments , etc. >>The advantage that you and me have is that we can try out all our ideas and >>release versions which incorporate the newer ideas. If they fail - too bad , >>back to earlier versions. >>For commercials - they have to make sure that nothing fails , there are no >>regressions - making sure that there are no regressions is a major painful >>expierence. >>Also , commercials will have schedules by which time they have to release a >>newer and more improved version .... this is an added pressure in this >>commercial arena that amateurs dont have. > >I do not think that they have these schedules. > >Christophe does not release new tiger for a long time because he failed to >improve it signficantly. > >>Considering that I reject lot of ideas after implementing and testing it for >>weeks , it is quiet easy to imagine the pain the commercials programmers go >>through !! > >For me the problem is not ideas but implementation. > >There are ideas that I feel sure that they are good but I did not implement them >because they did not help me probably because of a bug in the implementation. > >The problem is that I am not good in doing specific well defined tasks and this >is the reason that movei has no book that knows about tranposition. > >Another programmer could probably translate the books that I have for movei to a >book that knows about tranposition in few hours but it is not the case for me. > >I have 2 books one for white and one for black and I want to translate them to >one book that has only positions of white to move and move(every position with >black to move is equivalent to position with white to move and I want the >program to be in book not only after 1.e4 e5 but also after 1.e3 e5 2.e4 so it >means that I need to have also a function that translates position and move of >black to position and move of white). > >Of course I also want the program to be able to search efficiently to find the >move in the book(today movei does not search efficiently to find book moves but >it is not very important because the book is very small). > >I do not plan to include learning except the learning that I have today to >change the first move after a loss to avoid losing the same game twice in a >match of 4 games before I change the structure of the book to be in the way that >I want. > >Uri The comment "it takes me 6 months to get a speed improvement of .5%" does not mean, as Vincent suggested, that someone like Frans spent 6 months to get that .5%. It means "to get a .5% speed improvement is hard." Nothing more. It is a form of hyperbole used to make a point, by implying how difficult something is. To think that someone would actually spend 6 months for .5% is so far beyond ridiculous it takes sunlight 3 months to get from ridiculous to there...
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.