neural network problem in newff

eror shown was Error using newff (line 107) Input ranges is not a two column matrix. Error in concir1 (line 25) net=newff(input_n,target_n,12); >> wedinput.dat data has 24 * 98 and wedtarget.dat data has 24*1 double . matllab versionis 8.0.0.783 R2012B.

matlab version 10a gives the output for my collegaue. please help me in this regard.

Answer 1

[Greg Heath]

close all ; clc; clf; clear all;

% Load the Closing stock index Data

load wedinput.dat; load wedtarget.dat;

% Input and Target Data

input1=wedinput(1:24,1:198); target1=wedtarget(1:24,1);

1. INCORRECT. INPUT AND TARGETS SHOULD HAVE THE SAME NUMBER OF COLUMNS.

2. INPUT DIMENSIONALITY REDUCTION RECOMMENDED

% Preprocessing the input and target data % y = (ymax-ymin)*(x-xmin)/(xmax-xmin) + ymin;[input_n,input_n_struct]=mapminmax(input1'); [target_n,target_n_struct]=mapminmax(target1');

3. INEFFICIENT. TRY MAPSTD ON INPUT WITH TANSIG HIDDEN NODES. REMOVE OR MODIFY OUTLIERS BEFORE TRAINING. IF NOT CLASSIFIER, USE MAPSTD ON OUTPUT ALSO.

P.S. EASIER TO USE ZSCORE

 HELP ZSCORE
 DOC  ZSCORE

disp('Input versus target After Preprocessing this data used for NN training');

4. PLOT 198 INPUTS ?

%[input_n target_n]; % Stote MSE,RMSE,MAE,MAPE and THEIL'U value of Train/Val/T

5. SPELLING

net=newff(input_n,target_n,12);

6. minmax(input_n) FOR YOUR OBSOLETE VERSION?

7. WHY H = 12?

%training the network net. %net.trainParam.epochs=1 net.divideFcn='divideblock';

net.divideParam.trainRatio=80; net.divideParam.valRatio=10; net.divideParam.testRatio=10;

8. Divide above by 100. BUT WHY NOT ACCEPT DEFAULTS?

net.trainFcn='trainlm';

9. DELETE ABOVE. IS A DEFAULT.

    [net tr Y E]=train(net,input_n,target_n);

 10. NMSE = mse(E)/mean(var(target_n',1)
 11. R2 = 1-NMSE
 12. USE tr TO OBTAIN VALUES FOR TRN/VAL/TST

% Simulation disp('Simulation Output for the given input'); simoutput=sim(net,input_n);

13. REDUNDANT simoutput = Y

[input_n ; simoutput];

14. DELETE ABOVE LINE. USELESS FOR N = 198

15. I'M STOPPING HERE.

%Output Normalized Version

disp('Normalized Simulation Output'); Norm_input=mapminmax('reverse',input_n,input_n_struct); Norm_simoutput=mapminmax('reverse',simoutput',target_n_struct);

%[Norm_input Norm_simoutput]; [Norm_input]; [Norm_simoutput] [target1 Norm_simoutput]

% Calculating Mean Square Error

perfmse=mse(E); fprintf('Mean Square Error:%d \n',perfmse);