Author: Thorsten Czub
Date: 19:12:00 12/28/97
Go up one level in this thread
>I see a paradox here. An asymetrical search should not behave this way, >because as you said it has seen every move you can do to it. So it has >also seen any sac you may do. The problem is: 1. you let genius compute your move. It will play e4. 2. you force e4 and let it compute again. It now has black and considers on a move. You do this to be prepared to know what the opponent is doing. This time genius searches the opponents plies with it's much-pruning plies 1,3,5,7... Therefore it could oversee something . I was always very careful with genius main-lines and genius evaluations BECAUSE of the pruning-effects of its search. A program that does not like to play its own main-lines just because you force ONE ply to move out of it does not seem to be very reliable ! I will try to SHOW you some positions that make it clearer if you want, when I have more time. It is again very late and I will go sleep now. >>Many people always said: >>This could mean M2 = 2 plies "brute-force" with overall S15=15 plies >>selective peak. >> >>I don't think this interpretation was ever true. From my point of view M2 does not tell us that genius is 2 plies brute-force deep ! From my point of view it is 2 times of the brute-force-parts-deep !! That would imply: 1 selective 2. brute-force = 1st M, 3 selective 4. brute-frce = 2nd M. So in fact M2 is not 2 plies but instead: 4 ! This could also explain the strange matrix I have seen to the base of 2 ! The 2 is nothing else than the 2 different searches and this results in those long -main-lines. The main-line for M0 is ply1, ply2 The main-line for M1 is ply1, ply2, ply3 For m2 it is ply1, ply2, ply3, ply4, ply5 because ply 2 and 4 are 2 times brute-force plies it indicates M2. This way the M indicator would correlate much better to the length of the main-lines. I must admit that I don't know exactly if this is the pattern we analysed in paris in the hotel. If Dan Wolf , the Gandalf programmer or Vincent would help us a little, they could confirm or deny it. I was tired in Paris and it was late. But it was obvious that there was a formula to the power of 2 correlating the length of the mainlines and the search depth in a formula. >I'm not sure you need to go in such a strange field to explain the Lang >algorithms behaviour. But tell us more! :-) I am not a programmer. But I am good in seeing patterns. When I see a structure it takes not much time until I see the mechanism behind. If we misunderstood the indicators the old mephisto machines showed us about the search depth, we have a problem explaining WHY the main line is that long and accurate. If we have a "different" idea what these indicators show, we can maybe explain the long main-lines. Maybe this is just like a cross-word-secret. But it makes fun to crack ideas. >This could mean Genius uses a lot of static piece-square tables. >Programs using this approach often make serious errors when going from >middle game to endgame for example. This generally happens after a queen >exchange. See how much Genius evaluation drifts after such an exchange! This is - in fact - the problem with genius. It makes very very often mistakes i understanding how to evaluate the END of it's main-lines IF there are captures in it that lead to a different stage of the game, e.g. from mid- to endgame. Your reasonal explanation fits perfectly well to my observations. Thanks. >>Today his concept is rusty. It still works but cannot reach the top >>again. > >Wrong. It is a more intelligent approach, and could well be the future >of computer chess. IF his concept IS so good, why wasn't he able to make any progress since genius 3 ??? Or even genius3, is not much stronger than Genius2 = Genius3 without X-mode. There are not many differences in playing strength between the genius versions since genius2. Give genius2 an X-mode and you will see. Give Genius3 a learning function and a book that is NOT outcooked by the opponents like the Genius3 book, and you will see a genius3 overtake the genius5 elo in the ssdf-list. And - as I said, Genius3 was only a genius2 with X-mode. Again - when my memory does not exchange some old days.... long time ago.... > >But I think we don't speak about the same thing. You were talking about >the asymetrical search, and you are right in this field. I'm talking >about expanding lines at the end of the search. > >So amateur programmers opinions are welcome. Together, can we get into >Lang's secrets before censorship covers the subject? :) :-) I will give some positions for you the next moment I find time. I don't think horizont problems are the reason Genius is NOT finding the key-move there. I think the key-move is pruned away. And it is found very fast if you take bacl one move from the position of go forward one ply. You can imagine why. The asymmetric search causes the "overseen" problem. Maybe I find also positions where this search causes a passive line instead of a more active one... Good night so far...
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.