cmac1d.m

人工神经网络中cmac网络应用于机器人手臂控制的算法研究

function cmac1d()
%

% A single input single output CMAC network



help cmac1d
clf reset
pausetime = 0.1;


P = -1:0.1:1;



P3 = -1:0.005:1;

% INITIALIZE 

iprange = 256;
disp(sprintf('iprange is %d',iprange));

%width = input('enter generalisation width ');

%c = input('enter number of weights to be accessed, c ');

%beta = input('enter learning rate, beta ');
width=1;
c=10;
beta=0.5;
memreq = c*ceil(iprange/width);
s = sprintf('number of weights required is %d',memreq);
disp(s);

%memsize = input('enter memsize (0 to disable hashing) ');
memsize=0;
% INITIALISE WEIGHTS
% ==================

if memsize == 0
  wts = zeros(memreq,1);
else
  wts = zeros(memsize,1);
end

% PLOT CMAC OUTPUT


[wts,netout] = modcmac(wts,[(2.0)*64],1,1.0,1.0,iprange,c,width,memsize);

for k=1:401
  [wts,op(k)] = opcmac(wts,[(P3(k)+2.0)*64],1,iprange,c,width,memsize);
end;



plot(P3,op,'r');
axis([-1.0,1.0,-2.0,2.0]);
title('CMAC output after one training example');
xlabel('Input Vector P');
ylabel('Network Output');

disp('any key to continue');

pause


%Define the associated twenty-one 1-element targets.

z = menu('select target function', ...
  'damped sinusoid', ...
 'step', ...
'impulse');
disp('')

if z == 1
  T = [-.9602 -.5770 -.0729  .3771  .6405  .6600  .4609 ...
        .1336 -.2013 -.4344 -.5000 -.3930 -.1647  .0988 ...
        .3072  .3960  .3449  .1816 -.0312 -.2189 -.3201];
else
  if z == 2
   T = [0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1];
else
   T = [0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0];
end
end


plot(P,T,'+');
title('Training Vectors');
xlabel('Input Vector P');
ylabel('Target Vector T');
hold on

% PLOT INITIAL
if memsize == 0
  wts = zeros(memreq,1);
else
  wts = zeros(memsize,1);
end

for i=1:401
  [wts,op(i)] = opcmac(wts,[(P3(i)+2.0)*64],1,iprange,c,width,memsize);
end

plot(P3,op,'m');
title('Function Approximation');
xlabel('Input Vector P');
ylabel('Target + Network -');
h = get(gca,'Children');
h = h(1);
 hold on
pause2(pausetime);

% TRAIN THE NETWORK
%==================

% TRAINING PARAMETERS
max_epoch = input('enter maximum number of training epochs ');

epoch = 1;
SSE = 1;
while (epoch < max_epoch) %& (SSE > 0.02)

for i = 1:21
 [wts,netout] = modcmac(wts,[(P(i)+2.0)*64],1,T(i),beta,iprange,c,width,memsize);
end
for k=1:401
  [wts,op(k)] = opcmac(wts,[(P3(k)+2.0)*64],1,iprange,c,width,memsize);
end
%for i=1:memreq
 %   disp(sprintf('%d',i));
  %  disp(sprintf('wts is :%d',wts(i)));
  %end;
  
    for i = 1:21
      [wts,p]=opcmac(wts,[(P(i)+2.0)*64],1,iprange,c,width,memsize);
      neterr(i) = T(i) - p;
    end
    SSE = sumsqr(neterr);
    fprintf('epoch = %d  SSE = %5.3f\n',epoch,SSE);
    delete(h);
    h = plot(P3,op,'m');
    drawnow
    pause2(pausetime);
    epoch = epoch + 1;
    end


end

disp('any key to continue');

pause

hold off
bar(wts,'r')
title('Network Weights');
xlabel('Weight Number');
ylabel('Weight Value');
if memsize == 0
  set(gca,'xlim',[0 memreq]);
else
  set(gca,'xlim',[0 memsize]);
end



function [wts,op]=opcmac(wts,ip,ipdim,iprange,c,width,memsize)
% OP = OPCMAC(WTS,IP,IPRANGE,C,WIDTH,MEMSIZE)
HASH=12345;
addrs=zeros(c,1);
quantisation=ceil(iprange/width);
offset=width/c;
ofs=0;
op=0;
for i=1:c
    address=0;
    shift=1.0;
    for j=1:ipdim
       %address=address+rem((ceil((ip(j)+ofs)/width)),quantisation)*shift;
       address=address+ceil(ip(j)/width)*shift;
       shift=shift*quantisation;
    end;
   % address=address+shift*i;
    if memsize>0
        addrs(i)=ceil(rem(((log(address+1))*HASH),memsize));
    else
        %addrs(i)=ceil(rem(address,c*ceil(iprange/width))); 
        memreq = c*ceil(iprange/width);
        addrs(i)=ceil(rem(address,memreq));    
    end;
        ofs=ofs+offset;
        op=op+wts(addrs(i));
 end;


function [wts,mod]=modcmac(wts,ip,ipdim,target,beta,iprange,c,width,memsize)
  %sub function
  HASH=12345;
addrs=zeros(c,1);
quantisation=ceil(iprange/width);
offset=width/c;
ofs=0;
sum=0.0;
for i=1:c
    address=0;
    shift=1.0;
    for j=1:ipdim
        %address=address+rem((ceil((ip(j)+ofs)/width)),quantisation)*shift;
        address=address+ceil(ip(j)/width)*shift;
        shift=shift*quantisation;
        
    end;
    address=address+shift*i;
    if memsize>0
        addrs(i)=ceil(rem((log(address+1)*HASH),memsize));
    else
        %addrs(i)=ceil(rem(address,c*ceil(iprange/width))); 
        memreq = c*ceil(iprange/width);
        addrs(i)=ceil(rem(address,memreq));
    end;
        ofs=ofs+offset;
        sum=sum+wts((addrs(i)));
  end;
   delta=beta*(target-sum)/c;
   for i=1:c
       wts(addrs(i))=wts(addrs(i))+delta;
   end;
   mod=sum+delta*c;
  %s=sprintf('mod is %d',mod);
   %disp(s);
 %  for i=1:c
  %disp(sprintf('wts is %d',wts(addrs(i))));
  %disp(sprintf('addrs is %d',addrs(i)));
  %end;