Author: Dann Corbit
Date: 12:10:43 02/09/06
Go up one level in this thread
On February 09, 2006 at 14:47:21, Uri Blass wrote: >On February 09, 2006 at 14:24:49, Dann Corbit wrote: > >>On February 09, 2006 at 14:22:03, Uri Blass wrote: >> >>>On February 09, 2006 at 14:08:31, Dann Corbit wrote: >>> >>>>On February 08, 2006 at 22:17:11, Will Singleton wrote: >>>> >>>>>>The very rapid >>>>>>improvement in the general level of strength isn't a recent >>>>>>development; it has been going on for several years. The gap >>>>>>between the new and improving programs and the established >>>>>>professionals has been constantly diminishing, and it has long >>>>>>been clear that it was only a matter of time before some of the >>>>>>new engines would surpass the old giants. >>>>> >>>>>Uh huh. It might have long been clear to you, but was that before Fruit or >>>>>after? I recall when Ruffian came out, people could scarcely believe its >>>>>strength. No one, folks said, could develop such a strong amateur program in >>>>>relative secrecy, and then just burst on the scene. But Ruffian never >>>>>approached commercial strength. >>>>> >>>>>And then came Fruit. Not an evolutionary change, not forseen, not anticipated. >>>>>One day the commercials ruled, then Fruit came out. That was the change. Only >>>>>after that did programs begin to "surpass the old giants." I'm not criticizing, >>>>>only pointing out that the change was sudden, not evolutionary, and no one >>>>>predicted it. >>>>> >>>>>Unless, as you say, you did. Where was that post? >>>> >>>>Junior and Chess Tiger were just like Fruit. It's just history repeating >>>>itself. And it is going to happen again and again >>> >>>No >>>Fruit was free source code. >>>Junior and chess tiger were not free source code. >>> >>>It is not history repeating and you can see big advance in copmputer chess after >>>fruit2.1 >>> >>>In less than a year from fruit2.1's release we already got 100 elo improvement >>>only in software relative to shredder9 even if we only look at the 32 bit >>>version of rybka. >>> >>>When did it happen in the last 10 years? >>> >>>I think that in the last 10 years there was never improvement of 100 elo >>>relative to previous best program. >>> >>>3 Rybka 1.01 Beta 13-13b 32-bit 2870 28 28 488 74.0 % 2689 27.5 % >>>14 Shredder 9 2753 8 8 4805 63.2 % 2659 31.9 % >>> >>>Note that we can expect Rybka1.2 to be even stronger so we probably get >>>improvement of more than 150 elo in the last year. >>> >>>I think that it never happened in the last 10 years and maybe never happened in >>>the past. >> >>I guess that when programs went from Mini-max to Alpha-Beta, the improvement was >>at least 200 Elo on average. > >This was a long time ago and it is easier to make an improvement when the level >is weaker. > >I did not follow improvement in computer chess not in the last 10 year but I do >not remember big improvement like the improvement in the last year when you >compare programs on the hardware that people use. > >Note that most people do not use 2 processors so it is logical to make >comparison of performance on the hardware that is used. In this respect I do not think that chess programming is different from any other discipline. It is something that moves in jumps and starts, historically, and makes big advancement when a fresh new idea becomes known. In sorting, quicksort was a revolution. Also radix sort. There are some new and brilliant ideas like relaxed-heaps and things of that nature that continue to move ahead. For a single processor, there is a theoretical minimum amount of energy that must be expended to perform a comparison sort which is log(n!) but which can be circumvented by sorts not requiring comparisons. So for chess also, there may be some physical laws that limit a particular kind of attack from attaining perfect chess. All exponential problems are extremely difficult to solve, which is what I think is the most attractive thing to chess programmers. If chess programming were easy, we would not find 300 chess programs. There may be a fundamental breakthrough in chess programming like distribution based sorting compared to comparison based sorting. There may be algorithmic improvments like sorting going from O(n^2) to O(n^(5/3)) to O(n*log(n)) for insertion sort -> shell sort -> quick sort algorithm improvements. The alpha-beta improvement to minimax is such an idea which reduces the complexity of the problem a great deal. Perhaps there is something equally clever in the offing. Just because we have trouble imagining it does not mean that it cannot be done.
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.