Author: Thorsten Czub
Date: 07:09:14 06/06/03
Go up one level in this thread
On June 06, 2003 at 06:06:31, Sergei S. Markoff wrote: >Botvinnik was a person with a very "wide" mind (: thats true. But i think i understood what he meant. >I can show you some of ideas that can be used to construct working method. >For example, you can found that some chess operations includes the moving some >piece from one board region to another. It can be easily found for example in >several endgames. >Null move doing a very bad thing in some positions. It "cuts" some of this >trajectories. exactly. that is the reasoon null move is senseless IMO for knowledge programs. it cuts the PLANS out of the programs main-lines. >For example to finish attack on opponent king we must move rook >from f1 to g3. Right. that would be a trajectory. together with Queen d1-g7 mate :-))) So we have to move the queen (or other pieces) too to g7 :-)) the good idea of botwinnik was (IMO) that those trajectories are not legal moves but geometrical moves the human eye sees. those trajectories must later be FULLFILLED with legal moves of chess, the search must put those moves to get the trajectories in the right order. So in fact those chess programs would PLAN like humans do. that was (IMO) the main plot. >This operation includes, for exapmle, 3 plies: f1-e1, e1-e3, >e3-g3. We need to search with Depth = 6 plies to found this operation. If we has >one or tho pawns less before performing this operation, null-move can found that >when opponent not moving we, anyway, can't do something good in 4 plies (R=2) >and it causes cut-off. null moves cut the moves that are in charge for making the trajectorie come true. >The other problem is endgame. Is some pawn endgames we must move our king from, >say, h1 to e7. But our search is "blind". 12-ply search discovers right >trajectory, but it also discovers a lot of really stupid king movements. In some >cases it's a good idea to extend some trajectories if we can see a goal - some >pawn rams or something else. In SmarThink I tried to extend some nodes in which >we see that previous king movement is a shape with no self-intersections and >without moving to squares that can be reached by king early. This method also >need some special cut-offs to stop cathastropic tree expanding. > >I'm working to improve all of this methods. And I'm interesting in your ideas (; Great. This sounds interesting. I hope you can use those methods more. Todays chess programs need to PLAN. noot by random, but by WILL ! but instead of creating programs that DO plan, programmers create chess programs that search deep without a clue of a plan. they only evaluate positions, but not ideas ! they evaluate MOVES but not chess. Botwinnik saw that you have to evaluate a PLAN ! not aa move nor a position. Todays chess programs play like Tarrasch. they want to find the right move in a position. that is NOT chess. that is idiot-chess.
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.