带子网的极限学习机（分类或预测）附matlab代码.zip

function [Training_Accuracy,Testing_Accuracy]=elm_subnets(X,Y,Xtest,Ytest,number_neurons,number_subnets,neurons_subnets,ELM_Type) % elm_subnets:it is used to train a Single hidden Layer Feedforward Network with subnets based ELM ; % X=a training set; matrix of N instances by Q atrebutes ; % Y=a training targets; matrix of N raws and 1 atrebutes of the target values; % xtest= testing inputs ; % ytest= testing targets; % number_neurons:number of neurons in the hidden layer (user decision); % number_subnets:number of subnetworks in the hidden layer % neurons_subnets :number of neurons in each subnetwork in the hidden layer % ELM_Type: ELM_Type=1 for classification;ELM_Type= any Other values for regression; % Training_Accuracy:RMSE for regression and cllassification rate for classification. % A: a vector contains the idexes of subnets in the hidden layers % note: for classification you should spicify classes with integers and % each instance hase only one class(one target for each instance); %%%%%%%%%%%%%% %%%training%%% %%%%%%%%%%%%%% X=scaledata(X,0,1);% data normalization Xtest=scaledata(Xtest,0,1);% data normalization alpha=size(X);% dimension of training set. %1)first step:generate a random input weights input_weights=rand(number_neurons,alpha(2))*2-1;% generate random input weights %2) second step:calculate H matrix H=input_weights*X';% dot product H=radbas(H);% activation function(you have the ability to change you function here) %3) third step: subnets mapping if number_subnets<=number_neurons A=randi([1,number_neurons],1,number_subnets);% randomly shoose the position of subnets in the hidden layer end SUB_inputweights=rand(neurons_subnets,number_subnets)*2-1;% generate input weights for the subnets; SUB_outputweights=rand(number_subnets,neurons_subnets)*2-1;% generate output weights for the subnets; for i=1:number_subnets input_subnet(i,:)=H(A(i),:); % load the A th hidden neurons end H_subnet=SUB_inputweights*input_subnet;% temporel hidden layer for the subnet H_subnet=radbas(H_subnet);% activated hidden layer for the subnet O_subnet=(H_subnet' * SUB_outputweights')';% the output of the subnet O_subnet=radbas(O_subnet);% the activated output of the subnet for i=1:number_subnets H(A(i),:)=O_subnet(i,:); % update H end %%%%end subnets mapping %4)forth step : calculate the output weights Beta B=pinv(H') * Y ; %Moore-Penrose pseudoinverse of matrix %%%%%%%%%%%%%%%calculate the training prefomance %%%% if ELM_Type==1 tr_output=round(H' * B)' ;% calculate the training output c_rate=zeros(1,length(tr_output)); for tb2=1: length(tr_output) if tr_output(tb2)==Y(tb2) c_rate(tb2)=1; end end Training_Accuracy=sum(c_rate)*100/length(tr_output);%Classification rate else tr_output=(H' * B)' ;% calculate the training output Training_Accuracy=sqrt(mse(Y',tr_output));%rmse end %%%%%%%%%%%%%%%% %%%testing%%%%%% %%%%%%%%%%%%%%%% H_test=input_weights*Xtest';% calculating H matrix H_test=radbas(H_test);% activation function(you have the ability to change you function here) %%% subnets mapping for i=1:number_subnets ts_input_subnet(i,:)=H_test(A(i),:); % take samples end ts_H_subnet=SUB_inputweights* ts_input_subnet;% % temporel hidden layer for the subnet ts_H_subnet=radbas(ts_H_subnet);% the activated hidden layer for the subnet ts_O_subnet=(ts_H_subnet' * SUB_outputweights')';% the output the subnet ts_O_subnet=radbas(ts_O_subnet);%% the Activated output the subnet for i=1:number_subnets H_test(A(i),:)=ts_O_subnet(i,:); % update H_test end %%% end of subnets mapping %%%%%%%%%%%%%%%calculate the testing output %%%% if ELM_Type==1 ts_output=round(H_test' * B)' ;% calculate the training output else ts_output=(H_test' * B)' ;% calculate the training output end %%%%%%%%%%%%%%%%calculate the prefomance%%% if ELM_Type==1% calculate the output c_rate=zeros(1,length(ts_output)); for tb2=1: length(ts_output) if ts_output(tb2)==Ytest(tb2) ts_c_rate(tb2)=1; end end Testing_Accuracy=sum(ts_c_rate)*100/length(ts_output);%Classification rate else Testing_Accuracy=sqrt(mse(Ytest',ts_output));%rmse end end