BubbaTough wrote:I looked back through the thread a bit, and see this has already been rehashed. I have not looked through all of it though, so must have missed some I am sure (the thread is kind of long at this point).
OK, it is a good point many have made...different numbers are indeed generated. However, it seems like the same properties should hold right? I don't see any reason for alpha-beta to detract from the effects aimed for in any way...the nodes not expanded are known to have no effect if they are expanded, and while they change what number will be generated next by causing the pseudo-number generator to use a different seed...why would that be a bad thing? It just doesn't make sense to me that skipping superfluous random number generations should fundamentally effect the properties of the tree.
I apologize if this has already been covered and I am rehashing old areas. Feel free to ignore me, reference me to old posts, or call me stupid if desired.
-Sam
I don't yet understand minimax fully and less so what pure minimax is. Neither do I know much of what is a game of "perfect information". I try only to post what I seem to know as correct with much holes in knowledge and uses "intuition" (?) to fill the gaps.
Alphabeta should be sufficient in demonstrating the Beal effect - the Beal Effect is minimax search with "perfect knowledge". It is just a peculiar aspect of minimax and, maybe, the Beal effect is nature's way about "perfect knowledge" which is above "perfect information" - both players equally and absolutely have perfect knowledge when both have no knowledge.
Only a while ago did I know what is meant by "random numbers independent of positions" :-
1) a normal alphabeta returns a deterministic best pv barring subtleties.
2) alphabeta with a random evaluation from hash also returns a deterministic best pv.
3) alphabeta with a evaluation from rand() do not return a deterministic best pv. It may return any one of the legal moves as the best move and research don't give the same pv.
But both 2) and 3) should exhibit the Beal Effect and possibly with the same probabilistic properties.
I think it might be possible to analyze the Beal effect without knowing much what happen at the leaves but through bounds (alpha, beta) passed up and the best score passed down the tree. At any node with (alpha, beta) and N search scores to select from, the probability that the node is the new pv increases with N. It does not matter if this node fails low or fails high. The pv will exhibit the Beal Effect.
4) The score return from the root search probably could also be used as a random number generator.
5) The best move returned is NOT a random move selected randomly, but selected "intelligently" due to the Beal's effect.
Rasjid