twosgconstrained.m

来自「一个启发式搜索算法 给初学者用的 随便参考着用 提供了多种策略」· M 代码 · 共 17 行

% J.C. Spall, Sept. 1998
% twoSGconstrained
% Code for evaluation of second-order SGSA (2SGSA) versus first-order 
% SGSA (1SGSA, as in Chap. 5 of ISSO).  Code is for comparative evaluation purposes; hence,  
% it includes much that is not required for a basic implementation. Further, it is 
% in no way "optimized" for efficiency or generality; this is strictly research
% code for the purpose of getting a basic idea of how 2SGSA works.
%
% Code includes the capability for initializing 2SGSA by running 1SGSA for
% N iterations.  This code allows for averaging of the 
% SG gradients and Hessian estimates at EACH iteration after the 
% initial (N) iterations where only 1SGSA is used for estimating theta.  We use "thetaH"
% for the 2SGSA recursion and "theta" for the 1SGSA recursion.
%
% UPDATE MAR. 2006: Feedback and weighting versions of this code are available from the author;
% this can generally provide enhanced performance. (Reference: Spall, J. C. (2006), 揊eedback and Weighting Mechanisms 
% for Improving Jacobian (Hessian) Estimates in the Adaptive Simultaneous Perturbation Algorithm,