Running a validation model using Stepwisefit

I have produced the following code. It does 1505 stepwise regression with 8 predictor variables (a regression for every pixel of a map). As you can see it runs 2 regressions, 1 using all of the predictor variables and one using only predictor variables deemed to be in the model (INMODEL output from regression).

The regressions use data 1:127 from the input data.

Now I want to produce a validation set. Basically I want to run the same regression but using new data (128:144). It should only use variables deemed to be in the model in the first stepwise regression.

I am unsure of how to proceed. Any advice would be greatly appreciated

 %Sort the Data
    [rows, cols]=size(A11);

    % initialize matrix to store the data
    pvalues=zeros(rows, cols)-1;
    coefficients=zeros(rows,cols);
    for i = 1:rows
      for j = 1:cols
          y=zeros(8,1)
          x=zeros(8,8)
          for m=1:127                   %number of months
              y(m)=All((m-1)*rows+i,j);
              x(m,1)=Approx1((m-1)*rows+i,j);
              x(m,2)=Approx2((m-1)*rows+i,j);
              x(m,3)=Approx3((m-1)*rows+i,j);
              x(m,4)=Approx4((m-1)*rows+i,j);
              x(m,5)=Dpprox1((m-1)*rows+i,j);
              x(m,6)=Dpprox2((m-1)*rows+i,j);
              x(m,7)=Dpprox3((m-1)*rows+i,j);
              x(m,8)=Dpprox4((m-1)*rows+i,j);
              x(m,9)=1;
 end
        %Run the stepwise Regression
        if  sum(sum(isnan(x)),2)==0 && sum(isnan(y))==0 
        xx=x(:,1:9); yy=y(:,1);

        [B,SE,PVAL,INMODEL,STATS,NEXTSTEP,HISTORY]= ...
                                       stepwisefit(xx,yy,'penter',.05);

        inApprox1(i,j)=INMODEL(1);  %Find variables that are in the model
        inApprox2(i,j)=INMODEL(2);
        inApprox3(i,j)=INMODEL(3);
        inApprox4(i,j)=INMODEL(4);
        inDpprox7(i,j)=INMODEL(5);
        inDpprox8(i,j)=INMODEL(6);
        inDpprox9(i,j)=INMODEL(7);
        inDpprox10(i,j)=INMODEL(8);

        sstotApprox1(i,j)=STATS.SStotal;        %calculate R^2
        ssresidApprox1(i,j)=STATS.SSresid;
        rsq = 1- ssresidApprox1./sstotApprox1
        rsq(rsq==Inf) = NaN                     %Set Inf to NaN

        rmse(i,j)=STATS.rmse;              %Extract rmse
        rmse(rmse==Inf) = NaN           %Set Inf to NaN

        % repeat regresson only on the sigificant variables

        xip=0;
        for k=1:9   %9 refers to previous 9 variables including intecept
            if INMODEL(1,k)==1
                xip=xip+1;
                xxn(:,xip)=xx(:,k);
            end
        end
          [Bn,SEn,PVALn,INMODELn,STATSn,NEXTSTEPn,HISTORYn]= ...
                                       stepwisefit(xxn,yy,'penter',.05);      
                rmsen(i,j)=STATSn.rmse;              %Extract rmse
                rmsen(rmse==Inf) = NaN           %Set Inf to NaN
        end
    end
end

Answer 1

[Sruthi Ayloo]

Hi Robert,

I understand that you want to run the stepwise regression on your validation set using a different dataset (128:144). I see that your script has the following line where it takes in the input data:

for m =1:127 % number of months

One way to do is you can create a function which takes in an input and performs stepwise regression. This input can either be the train data or the validation data or the test data and this function can be called in your script. You can either declare the variables deemed to be in the model in the first stepwise regression as global so that they can be used in the function or you can return them as function output values.

Scripts Vs Functions discusses the differences between scripts and functions and shows how to convert a script to a function.

Functions - has more information on how to declare functions and call them.

Hope this helps.