newstsr.zip_grassbqh_一种神经网络构造的新方法

function net = jordan_p1(p,t,s,tf,btf,blf,pf,ipf,tpf,ddf) if nargin < 1, error('NNET:Arguments','Not enough arguments.'),end % Defaults if (nargin < 3), s = []; end if (nargin < 4), tf = {}; end if (nargin < 5), btf = 'traingdx'; end if (nargin < 6), blf = 'learngdm'; end if (nargin < 7), pf = 'mse'; end if (nargin < 8), ipf = {'fixunknowns','removeconstantrows','mapminmax'}; end if (nargin < 9), tpf = {'removeconstantrows','mapminmax'}; end if (nargin < 10), ddf = 'dividerand'; end % Error checking if (~isa(p,'double')) || ~isreal(p) error('NNET:Arguments','Inputs are not a matrix or cell array with a single matrix.') end if isa(s,'cell') if (size(s,1) ~= 1) error('NNET:Arguments','Layer sizes is not a row vector of positive integers.') end for i=1:length(s) si = s{i}; if ~isa(si,'double') || ~isreal(si) || any(size(si) ~= 1) || any(si<1) || any(round(si) ~= si) error('NNET:Arguments','Layer sizes is not a row vector of positive integers.') end end s = cell2mat(s); end if (~isa(s,'double')) || ~isreal(s) || (size(s,1) ~= 1) || any(s<1) || any(round(s) ~= s) error('NNET:Arguments','Layer sizes is not a row vector of positive integers.') end % Simulation net.inputs{1}.exampleInput = p; net.inputs{1}.processFcns = ipf; %pre-processing conditions for less arguments in the newjor for i=1:Nl if (i<Nl) net.layerWeights{i,i+1}.delays = 1; else net.layerWeights{Nl,Nl}.delays = 1 ; end if (i<Nl), net.layers{i}.size = s(i); end if (length(tf) < i) || all(isnan(tf{i})) if (i<Nl) net.layers{i}.transferFcn = 'tansig'; else net.layers{i}.transferFcn = 'purelin'; end else net.layers{i}.transferFcn = tf{i}; end end net.outputs{Nl}.exampleOutput = t; net.outputs{Nl}.processFcns = tpf; % Adaption net.adaptfcn = 'trains'; net.inputWeights{1,1}.learnFcn = blf; for i=1:Nl net.biases{i}.learnFcn = blf; net.layerWeights{i,:}.learnFcn = blf; end % Training net.trainfcn = btf; net.dividefcn = ddf; net.performFcn = pf; net.gradientFcn = feval(btf,'gdefaults',0); % Initialization net.initFcn = 'initlay'; for i=1:Nl net.layers{i}.initFcn = 'initnw'; end net = init(net); % Plots net.plotFcns = {'plotperform','plottrainstate'}; %=========================================== %this is similar to the above function written for the lower end MATLAB %version of NNT tooblox function net = jordan_p0(pr,s,tf,btf,blf,pf) % Backward compatible to NNT 5.0 if nargin < 1, error('NNET:Arguments','Not enough arguments.'),end % Defaults Nl = length(s); if nargin < 3, tf = {'tansig'}; tf = tf(ones(1,Nl)); end if nargin < 4, btf = 'traingdx'; end if nargin < 5, blf = 'learngdm'; end if nargin < 6, pf = 'mse'; end % Initialization net.initFcn = 'initlay'; for i=1:Nl net.layers{i}.initFcn = 'initnw'; end net = init(net); % Plots net.plotFcns = {'plotperform','plottrainstate'};