Author: Ilya P. Kozachenko
Date: 22:48:51 08/04/98
Go up one level in this thread
On August 04, 1998 at 20:25:43, blass uri wrote: > >On August 04, 1998 at 20:03:51, Fernando Villegas wrote: > >>Hi KK: >>I have posted about this very same issue some time, but now I wonder if inside >>this approach maybe there is a kind of hideous infinite progression. When the >>moment to decide which of the available evals must be used, tacitly we are >>thinking in a third engine or device to choose, to decide, but then which will >>be the eval of this third device to choose? ¿Positional, tactical, what? Then >>maybe we'll need a fourth device to give an alternative way to decide about the >>first lines of elvals and, and and. To avoid that circularity or progression the >>third device should be so crude than, then, all the idea lose ground. >>fernando > >There is a simple idea to decide. >start to use fritz5. >If you see fritz5 does not reach a big depth in a short time than stop using >fritz and use another program. >I think this idea will give a program that is better than fritz > >Uri This is what I'm waiting for. The key moment of this idea is, of course, choice of engine, i.e. type of position. It was made in Fritz5 GUI, but there remain questions: 1. How (good) was it made ? 2. Did anybody test it for adequate time ? There're 2 different approaches, I suppose, of implementation: a. Once decide which type position has (at root) and apply corr. algorithm; b. Use different methods during the search deepening (more complex); I'll consider only 1st approach here and leave 2nd for guru :) Since we make decision about position type only once, we'll not have any "hideous infinite progression". The most simple decision of position kind proposed Uri: 1. Look for speed of depth growing. I think it's possible to invent another methods (ask GMs :) but it is not a problem. Let's consider the BIG question: "Would it be better than each program alone ?" from logical point of view. Let's denote Positional Engine as PE and Tactical Engine as TE. Under these conditions it's possible to define 3 types of moves: 1. best move 2. good move (doesn't worsen position) 3. bad move (worsen position) Under these conditions it's possible to define 8 types of positions: 1. CP (Complex Positions) - both PE and TE can't find best move 2. VCP (Very Complex Positions) - both PE and TE can't find good move 3. TPP (Tactical-Positional Positions) - TE finds best move and PE finds good move 4. PTP (Positional-Tactical Positions) - PE finds best move and TE finds good move 5. TP (Tactical Positions) - TE finds best or good move but PE finds bad move 6. PP (Positional Positions) - PE finds best or good move but TE finds bad move 7. SP (Simple Positions) - both PE and TE can find at least good move 8. VSP (Very Simple Positions) - both PE and TE can find best move Wrong decision becomes appreciable only in TP and PP, while right decision in TP, PP, TPP and PTP. Since unit weight of TP and PP is quite small in comparison with TPP and PTP (I think, may be I'm wrong, I haven't sufficient experience, correct me if so), even if Decision Engine gives right choices in 50% of positions, it seems be nice. P.S. I wonder why gurus keep silence :( They have to have learned this idea far time ago. If it failes, please inform me for saving a lot of time of testing. Ilya.
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.