感知器算法Matlab源码

% bowl -- 3D display of error surface in weight space for LMS learning. load LmsPat.dat; load LmsAns.dat; Patterns = LmsPat'; Desired = LmsAns'; NINPUTS = size(Patterns,1); NPATS = size(Patterns,2); NUNITS = size(Desired,1); Inputs = [ones(1,NPATS); Patterns]; InitWeights = rand(NUNITS,1+NINPUTS)*2 - 1.5; % bias weights for larger error Weights = InitWeights; LearnRate = 0.142; % any larger than this, and the weights diverge colordef none, clf reset axpat=subplot(1,2,2); hold on PlotLmsPats(Patterns,Desired); % Calculate and plot the error surface fprintf('Calculating the error surface...'); w0range = -0.5:0.025:1.0; w1range = -0.5:0.05:2.5; w0range = -1:0.025:1.0; w1range = -1:0.05:3; [xm,ym] = meshgrid(w0range,w1range); zm = 0*xm; W = Weights; for i = 1:size(xm,1) for j = 1:size(ym,2) W(1,1:2) = [xm(i,j) ym(i,j)]; zm(i,j) = tss(W,Inputs,Desired); end end disp(''); axbowl = subplot(1,2,1); contour3(xm,ym,zm,50,'b'); hold on; axis([w0range(1) w0range(end) w1range(1) w1range(end) 0 max(zm(:))]) view (60,14); %view(220,50); grid off, box on xlabel('w0'); ylabel('w1'); zlabel('TSS'); title('Error Surface') set(gca,'CameraViewAngleMode','manual') rotate3d off, rotate3d on prevW = Weights; prevTSS = tss(prevW,Inputs,Desired); plot3(prevW(1,1),prevW(1,2),prevTSS,'or'); drawnow TSS = prevTSS; oldTSS = Inf; % Main loop of the LMS algorithm for i = 1:500, if rem(i-1,9) == 0 subplot(axpat) PlotLmsFn(Weights); subplot(axbowl); plot3([prevW(1,1) Weights(1,1)],[prevW(1,2) Weights(1,2)],[prevTSS TSS],'-r'); plot3(Weights(1,1),Weights(1,2),TSS,'or'); pause(0.5) prevW = Weights; prevTSS = TSS; end NetIn = Weights * Inputs; Result = NetIn; Error = Result - Desired; TSS = sum(sum(Error.^2)); fprintf('Epoch %.0f: ',i); fprintf('TSS = %6.5f\n',TSS); if abs(TSS-oldTSS) < 0.0001 subplot(axpat); PlotLmsPats(Patterns,Desired,1); PlotLmsFn(Weights,[1 1 1]); break; end; oldTSS = TSS; dW = - Error * Inputs'; Weights = Weights + LearnRate * dW; end