Author: Georg v. Zimmermann
Date: 03:36:29 10/22/00
Go up one level in this thread
On October 21, 2000 at 23:37:52, Pham Minh Tri wrote: >On October 21, 2000 at 17:55:22, Georg v. Zimmermann wrote: > >>On October 21, 2000 at 09:11:12, Pham Minh Tri wrote: >> >>>My questions are: >>>- If I add more knowledge to the evaluation function, how does AlphaBeta >>>function change (will the number of searched nodes increase, decrease or not >>>change)? >> >>If you add more knowledge, your program will probably spend more time in the >>eval() function. That way taking longer every time at the "leaf" of the search >>tree the evaluation is called. Threfore you will serach less nodes. >> >>How much this will make your program slower depends on what time - percentage it >>spends in which parts of the program before your change. If you have a very fast >>evaluation and slow bitboard move generation adding to the eval() will not cost >>you much. >> >> >>>- If I do not sort or sort move list by other criteria, could score and move >>>results be different? >> >>No, you should get exactly the same scores and moves after finishing a certain >>depth. The last move of the PV could change though in certain cases if I am >>making no thinking mistake. >> >>> >>>Your answers will help my implementation much. Thanks in advance for any help. >>> >>>Pham >> >> >>Have fun with your program ! >> >>Georg > >Thanks for your answer. However, we misunderstand some aspects of questions: > >1) I know that adding more knowledge makes searching slower, but how about the >number of searched nodes (without attention to search time) when there are more >"resolution" in the evaluation? You mean with fixed depth ? A bit it depends on how "exact" your evaluation is, eg. if you only return values that differ 0.1 pawn units you will search less nodes than with returning values which differ in centi-pawns. It also depends on how "consistant" your evaluation is, for example if mobility is important in your evaluation and you use so called "developement tables" too which give a bonus for knights on f3 and c3 for example you probably need to search a bit less nodes to reach the same depth. > >2) I think the resulted move could be diffirent when there are some of the best >moves (equal score). And I wonder if nullmove, razor and so on could affect the >resulted score when search finds another best move. Lets say in one position only 2 moves are possible and lead to mate in 5. Up until ply 9 (assuming plain alpha-beta) the PV-move depends on eval() of the final position, so the move that is searched first with 10ply does *not* depend on move ordering. Whatever move order, the same of both equal moves leading to mate will be picked. I hope I understood this time :) and I also hope someone will correct me if I am wrong. Georg
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.