hsx3.m

此程序是用于模式分类的在matlabhu环境下的原代码

function [b1,w1,w2,b2,w3,b3,i,tr,a3] = hsx3(n,m,f1,f2,f3,p,t,tp)
%TBPX2 Train 2-layer feed-forward network w/fast backpropagation.
%  
%  This function is obselete.
%  Use NNT2FF and TRAIN to update and train your network.

nntobsf('tbpx2','Use NNT2FF and TRAIN to update and train your network.')

%  [W1,B1,W2,B2,TE,TR] = TBPX2(W1,B2,F1,W1,B1,F2,P,T,TP)
%    Wi - Weight matrix for the ith layer.
%    Bi - Bias vector for the ith layer.
%    Fi - Transfer function (string) for the ith layer.
%    P  - RxQ matrix of input vectors.
%    T  - SxQ matrix of target vectors.
%    TP - Training parameters (optional).
%  Returns:
%    Wi - new weights.
%    Bi - new biases.
%    TE - the actual number of epochs trained.
%    TR - training record: [row of errors]
%  
%  Training parameters are:
%    TP(1) - Epochs between updating display, default = 25.
%    TP(2) - Maximum number of epochs to train, default = 1000.
%    TP(3) - Sum-squared error goal, default = 0.02.
%    TP(4) - Learning rate, 0.01.
%    TP(5) - Learning rate increase, default = 1.05.
%    TP(6) - Learning rate decrease, default = 0.7.
%    TP(7) - Momentum constant, default = 0.9.
%    TP(8) - Maximum error ratio, default = 1.04.
%  Missing parameters and NaN's are replaced with defaults.

% Mark Beale, 1-31-92
% Revised 12-15-93, MB
% Copyright 1992-2002 The MathWorks, Inc.
% $Revision: 1.12 $

if nargin < 7,error('Not enough arguments.');end

% TRAINING PARAMETERS
if nargin == 7, tp = []; end
tp = nndef(tp,[25 1000 1e-20 0.01 1.05 0.7 0.9 1.04]);%%##
df = tp(1);
me = tp(2);
eg = tp(3);
lr = tp(4);
im = tp(5);
dm = tp(6);
mc = tp(7);
er = tp(8);
df1=feval(f1,'delta');
df2 = feval(f2,'delta');
df3 = feval(f3,'delta');

dw1 = 0;
dw2 = 0;
db1=0;
db2 = 0;
dw3 = 0;
db3 = 0;
MC = 0;
%tp
%pause

[r,q] = size(p);
[s,q] = size(t);
b1=rands(n);
w1=rands(n,r);
w2= rands(m,n);
b2=rands(m);
w3=rands(s,m);
b3=rands(s);


% PRESENTATION PHASE
a1 = feval(f1,w1*p,b1);
a2 = feval(f2,w2*a1,b2);
a3 = feval(f3,w3*a2,b3);
e = t-a3;
SSE = sumsqr(e);

% TRAINING RECORD
tr = zeros(2,me+1);
tr(1:2,1) = [SSE; lr];

% PLOTTING FLAG

plottype = (max(r,s) == 1);

% PLOTTING
newplot;
message = sprintf('TRAINBPX: %%g/%g epochs, lr = %%g, SSE = %%g.\n',me);
fprintf(message,0,lr,SSE)
if plottype
  h = plotfa(p,t,p,a3);
else
  h = plottr(tr(1:2,1),eg);
end

% BACKPROPAGATION PHASE
d3 = feval(df3,a3,e);
d2 = feval(df2,a2,d3,w3);
d1 = feval(df1,a1,d2,w2);

for i=1:me

  % CHECK PHASE
  if SSE < eg, i=i-1; break, end

  % LEARNING PHASE
  [dw1,db1] = learnbm(p,d1,lr,MC,dw1,db1);
  [dw2,db2] = learnbm(a1,d2,lr,MC,dw2,db2);
  [dw3,db3] = learnbm(a2,d3,lr,MC,dw3,db3);
  MC = mc;
  new_w1 = w1 + dw1;
  new_b1 = b1+db1;
  new_w2 = w2 + dw2; new_b2 = b2 + db2;
  new_w3 = w3 + dw3; new_b3 = b3 + db3;

  % PRESENTATION PHASE

  new_a1 = feval(f1,new_w1*p,new_b1);
  new_a2 = feval(f2,new_w2*new_a1,new_b2);
  new_a3 = feval(f3,new_w3*new_a2,new_b3);
  new_e = t - new_a3;
  new_SSE = sumsqr(new_e);

  % MOMENTUM & ADAPTIVE LEARNING RATE PHASE
  if (new_SSE > SSE*er)
    lr = lr * dm;
    MC = 0;
  else
    if new_SSE < SSE
      lr = lr * im;
    end
      if lr>=1
          lr = lr * dm;
      end
      
    w1 = new_w1;
    w2 = new_w2; b2 = new_b2; a2 = new_a2;
    w3 = new_w3; b3 = new_b3; a3 = new_a3;
    b1 = new_b1;  a1 = new_a1;
    e = new_e; SSE = new_SSE;
    
    % BACKPROPAGATION PHASE
    d3 = feval(df3,a3,e);
    d2 = feval(df2,a2,d3,w3);
    d1 = feval(df1,a1,d2,w2);
  end

  % TRAINING RECORD
  tr(:,i+1) = [SSE; lr];

  % PLOTTING
  if rem(i,df) == 0
    fprintf(message,i,lr*.1,SSE)
    if plottype
      delete(h);
      h = plot(p,a3);
    drawnow
    else
      h = plottr(tr(1:2,1:(i+1)),eg,h);
    end
  end
end

% TRAINING RECORD
tr = tr(1:2,1:(i+1));

% PLOTTING
if rem(i,df) ~= 0
  fprintf(message,i,lr,SSE)
  if plottype
    delete(h);
    plot(p,a3);
  drawnow
  else
    plottr(tr,eg,h);
  end
end

% WARNINGS
if SSE > eg
  disp(' ')
  disp('TRAINBPX: Network error did not reach the error goal.')
  disp('  Further training may be necessary, or try different')
  disp('  initial weights and biases and/or more hidden neurons.')
  disp(' ')
end