Author: Martin Giepmans
Date: 03:21:46 08/03/01
Go up one level in this thread
On August 02, 2001 at 21:43:06, Miguel A. Ballicora wrote: >On August 02, 2001 at 19:45:26, Martin Giepmans wrote: > >>If the OS behaves as it should there is no randomness caused by other programs, >>interrupts or whatever. >>And if your compiler behaves as it should the rnd-function will always generate >>the same numbers (provided you use the same seed). >>In fact these numbers are not random at all. The rnd-function only imitates >>randomness. >>Of course you can introduce "randomness" the hard(ware) way, by kicking your >>computer, but there is really no soft way :) >>Randomness is a always a sign of a hardware-problem. >> >>Maybe I should explain what I have written earlier about node-counts. >>Suppose your program produces a weird move. OK, you count the nodes, it >>calculated 1888888 nodes, say. Now you set up the same position and you tell >>your program that it should calculate the same number of nodes, no less, no >>more, 1888888, period. (I this your idea? Or did I understand you wrong?) > >Yes > > >>Will your program play the same move? Will the score be the same? >>Nobody knows, because different information in the hashtables (etc) gives a >>different search-pattern (if this where not true then why would we use those >>tables in the first place?) > >The idea is that before each move the contents of the tables should be the same. >At the beginning of the game they are empty. In the first move, >I calculate the same amount of nodes and I reproduce the content of the tables. >Before the second move, I calculate the same amount of nodes starting with the >same tables so I reproduce the content of the tables after the second move and >so on... until the end of the game. > >Regards, >Miguel > Yes, this works. You will get the same game (same moves, same scores). But there is a problem. Suppose we find the bug that caused our famous weird move (24. Na1, say). We fix the bug and replay the game. Now what happens? Because we have changed our engine it will behave differently, not only in the position in which it moved Na1 but also *earlier* in the game. It may play 17.Rc1 instead of 17.Bh6 and we will never reach the position in which our weird thing happened... Even if the moves up to 24.Na1 are the same the content of the tables will probably be different because the engine has followed a different pattern. So if it now plays the brilliant Rxd1! instead of Na1? there might still be a bug... > >>Even the depth might not be the same: >>The second time it starts from scrap, nothing in the hashtables, no killers, >>etc, so it may need more then 1888888 nodes to complete depth 9, while the first >>time (when it produced the weird move) it needed exactly 1888888 nodes to >>complete depth 10.
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.