Author: Ratko V Tomic
Date: 10:18:30 10/23/00
Go up one level in this thread
> Bob Hyatt did not say that gambittiger is result of changing some > numbers like in chessmaster. ... > I suggested that a different evaluation function is the reason for > the fact that gambittiger sacrifies and has good results. I assume you're using term "evaluation" function in the sense of some computation done at each leaf node to estimate the node value, a function called at the leaf which returns a value of the leaf. (Otherwise, with a broad enough definition of "evaluation function" your and Bob's hypothesis can be claimed correct by definition.) With the assumed definition, and taking your explanation for what kind of novelty GT has, the GT and regular Tiger could be trivially made into one program where changing a number or two (the weights of the new and the old king safety evaluation term) you can switch from pure GT to pure CT and any point in between. There is no need to plan to integrate them in the future. Why would they ship two separate programs and why would Christophe mention that unifying the two programs will require quite a bit of work (well beyond cutting and pasting an evaluation function from one to another). Since he does wish (and intends) to unify them, it can't be a marketing gimmick to make it look as if 4 engines are in the distribution package, instead of 3. From the fact that it is a different program with the unification work planned in the future, I conclude that it is more than a new leaf node evaluator function for king safety. Since Christophe had mentioned that it wasn't a new kind of search algorithm, in order to still be non-trivial to transplant it into regular Tiger, there must be some more spread out type of new code, permeating the search routines and intertwined in a non-trivial way with the pre / post-processing logic. I suppose, Cristophe might be able to set this straight without blowing away any trade secrets. In any case, it ought to become clearer when we all had a chance to play with it a bit, run it through test positions and compare its nodes, times, depths & lines with regular Tiger.
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.