Author: Bob Durrett
Date: 05:52:41 12/06/02
Go up one level in this thread
Ingo: Your discussion of tactical versus positional is very insightful. Your ideas are closer to the problems chess programmers actually face when trying to make their engines do well with non-tactical positions. I would like to see more information about the different programming methods chess programmers have used to make their engines do well in non-tactical positions. Your discussion gives some insight in your approach. Also, your comments may be useful to programmers who are looking for a better way to solve this problem. Bob D. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ On December 06, 2002 at 04:18:00, Ingo Lindam wrote: >On December 05, 2002 at 20:31:41, Bob Durrett wrote: > >Hi Bob, > >I guess I will leave your 'GMs-in-a-box'-setting a little, when I say that for >me to distinguish between "positional" and "tactical" has mainly nothing to do >with the position, but with the way I make my decision about the best move or >just the evaluation of the current position. > >'Tactical' (for me) means to argue on basis of concrete variations and >'positional' on basis of (positional) features or pattern in the current >position, the experiences with this kind of features,... You can also put that >decissions based on exact knowledge into this class (of positional decissions) >or give trhem a third class (exact knowledge about a position is really rare, >but e.g. with help of EGTB you have it atleast for some positions, but also e.g. >a mate position you can exactly evaluate). > >So 'tactical' decissions are always based on 'knowledge based' >decissions/evaluations or 'positional based' decisions at the leaves of the >search tree. > >Playing games you will not often make pure 'positional' decisions, but also make >some 'tactical' thinking. So both will mix up in practical chess / playing games >usally. So to when seperating everything in just this two classes, you will have >you weight whether a decission it is more tactical or more positional. > >Just looking at a move that is played you can't say whether the decission is >made 'tactical' or 'positional'. Also to sacrify a Bishop on h7 is not necessary >(more) tactical in my opinion. You could look at the position and say there >enough pieces attacking the black kingside and too less black pieces defending >the king that I am sure, the king will be mated or atleast I will win enough >material by this attack (or atleast a perpetual check will be reachable). > >You ofcourse could classify the positions whether there is a proofable tactical >solution of it. Where you can proof such a solution the position could be called >(more) tactical. But when you don't see a proof within a certain scope a >position and you would like to call it a 'positional one' it might also be >'tactical', but you just need a deeper look into the search tree to see that. > >I guess you see that with my definition to distinguish between positional and >tactical it strongly depends on who plays or makes the evaluation of a position >or the decision for next move. So between human beings and the chess engines >today there should be a natural difference, because of the different scope. > >And you can obviously find some positions that you can claim to be tactical. But >to decide it is not tactical you have also to give a scope within that it is not >tactical. Assume you have a full tree (just theoretically assume) then nothing >positional will be left... everything turns into tactical proofable. > >So every position you name to be positional is just as long positional as you be >not able to proof the tactical solution. Again: Every so called 'positional' >position has a tactical solution. Just sometimes it is hidden too deep in the >tree. > >Internette Gruesse, >Ingo > > > > >> >>Imagine the following experiment: >> >>Several very strong chess players, preferably GMs, are put in a room and given >>the task of categorizing a very large set of chess positions. Assume also that >>they are willing participants. >> >>There is one large box filled with a large number of slips of paper. Each slip >>will have a single chess position on it. No two slips of paper have the same >>position. >> >>To obtain the positions on those slips of paper, a large number of positions >>were previously selected, using some *completely* random process, from a large >>database of high quality games. Being completely random, this process should >>have produced roughly equal numbers of opening, middlegame, and endgame >>positions. >> >>There are three large empty boxes, open at the top. One box is labeled >>"tactical." Another box is labeled "positional." The third box is labeled "no >>consensus reached within allotted time." >> >>The GMs are to try to reach a consensus. [No fair cheating!] They have a >>reasonable amount of time to examine each position. All positions for which >>consensus is not reached go in the box labeled "no consensus reached within >>allotted time." Otherwise, if the consensus is that the position is "tactical," >>the slip of paper goes into the box labeled "tactical." Finally, if consensus >>is reached that a position is "positional," it goes into the box labeled >>"positional." >> >>Would it not then reasonable to DEFINE positions in the "tactical" box to be >>"tactical" and the positions in the "positional" box to be "positional"? I >>believe that would be reasonable. >> >>I expect that, when all the slips had been examined, there would then be quite a >>few slips in the "no consensus reached within allotted time" box. My >>expectation is based on my perception that there are many positions that are, at >>the same time, both "tactical" and "positional," or maybe halfway in between. >> >>What does this have to do with computer chess? >> >>A second experiment could then be performed. Each position in the "tactical" >>box would be examined, this time, not by HUMANs but by the top chess engines. >> >>The purpose of the experiment would be to find tactical positions that presented >>problems for the chess engines, requiring too much time for evaluation. >> >>The same could be done for the positional and indeterminate positions. >> >>What would be the findings of this experiment? >> >>If modern chess engines are good at evaluating tactical positions, then there >>may be only a very few tactical positions found to be difficult for chess >>computers. >> >>If chess engines are poor at evaluating positional positions, then there may >>turn out to be a large number of positions found to be difficult for chess >>computers. >> >>And what about the positions taken from the "no consensus reached within >>allotted time" box? Your guess is as good as mine. >> >>Now, here is a test: [You have five seconds to answer.] What are the >>definitions of tactical and positional positions? [Tick, tick, tick.] >> >>Bob D.
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.