Author: Ricardo Gibert
Date: 16:31:58 07/07/03
I was reading about alternatives to alpha-beta search and got an idea that generalizes ETC (Enhanced Transposition Cut-offs). It is based on the following observations: 1 - The greater the draft of a cut node, the more confidence that the cut node will remain a cut node with deeper search. 2 - Having more than one move that produces a cut gives more confidence that the cut node will remain a cut node with deeper search. One problem with (2) is that additional cut moves may simply transpose into one another without having any real independent significance. However, if the backed up scores of each of the cut moves are not equal to one another, they can be presumed to be "independent". I'm not 100% sure about this, so let me know if this is reasonable. Now the way to generalize ETC based on (1) and (2) is to also accept "independent" transposing moves with a lower associated draft provided there are more than one of them. Such transpositions can be found at nearly zero cost, since when you fail to find a transposition of sufficient draft, you must run through all the legal moves anyway in your search for one. This can be even further extended by accepting even shallower drafts provided enough of them are found to offset risk. An example would be to accept two independent moves of draft = x - 1 in lieu of one move of draft x. If this is too risky, it could be made safer by requiring three moves of draft x - 1, etc. In a nut shell, the idea is to allow the trade off of draft for redundancy to increase the payoff of using ETC. If this idea is workable I would not expect it to give a big payoff. Also, it probably isn't original either, since it is rather simple.
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.