learnbm.m

来自「此程序是用于模式分类的在matlabhu环境下的原代码」· M 代码 · 共 35 行

function [dw,db] = learnbm(p,d,lr,mc,dw,db)
%LEARNBPM Backpropagation learning rule with momentum.
%  
%  This function is obselete.
%  Use LEARNGDM to calculate new weights and biases.

nntobsf('learnbpm','Use LEARNGDM to calculate new weights and biases.')

%  [dW,dB] = LEARNBM(P,D,LR,MC,dW,dB)
%    P  - RxQ matrix of input vectors.
%    D  - SxQ matrix of error vectors.
%    lr - the learning rate.
%    mc - momentum constant.
%    dW - SxR weight change matrix.
%    dB - Sx1 bias change vector (optional).
%  Returns:
%    dW - a new weight change matrix.
%    dB - a new bias change vector (optional).
%  
%  See also NNLEARN, BACKPROP, SIMFF, INITFF, TRAINBPX.

% Mark Beale, 1-31-92
% Revised 12-15-93, MB
% Copyright 1992-2002 The MathWorks, Inc.
% $Revision: 1.11 $  $Date: 2002/04/14 21:14:20 $

if nargin < 5,error('Not enough input arguments'),end
inc=lr*50;
x = (1-mc)*inc*d;
dw = mc*dw + x*p';
if nargout == 2
  [R,Q] = size(p);
  db = mc*db + x*ones(Q,1);
end