Convolution Neural network for regression problems

Hi everyone

I want to use CNN for my problem. The existing examples in the MATLAB (Here) provided for images as 4-D arrays but my problem is as follows:

Inputs = N (78000,24)

Output = Y(78000,1)

How can I use the mentioned examples for my problem?

Thanks in advanced.

[KSSV]

You can use NN toolbox right? Attach your data and tell us about your data, lets give a try to help you.

Answer 1

[yanqi liu]

yes，sir，may be use rand data to simulate your application，then you can replace data，such as

clc; clear all; close all;
% Inputs = N (78000,24);
% Output = Y(78000,1);
Inputs = randn(78000,24);
Output = rand(78000,1);
% get input data matrix
XTrain=(reshape(Inputs', [24,1,1,78000]));
YTrain=Output;

layers = [imageInputLayer([24 1 1])
    convolution2dLayer([15 1],3,'Stride',1)
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2,'Padding',[0 0 0 1])
    dropoutLayer
    fullyConnectedLayer(1)
    regressionLayer];

miniBatchSize  = 128;
options = trainingOptions('sgdm', ...
    'MiniBatchSize',miniBatchSize, ...
    'MaxEpochs',30, ...
    'InitialLearnRate',1e-3, ...
    'LearnRateSchedule','piecewise', ...
    'LearnRateDropFactor',0.1, ...
    'LearnRateDropPeriod',20, ...
    'Shuffle','every-epoch', ...
    'Plots','training-progress', ...
    'Verbose',false);
net = trainNetwork(XTrain,YTrain,layers,options);

[Jahetbe]

Thank you for your help.

Best regards,

Isa

[Jahetbe]

Dear @yanqi liu

Thank you for your answered.

Could you please help me to improve the accuracy of model?

I cannot find any optimum stduture to find my data not only when considered data for training and validations, but also when considered all of them for training.

Regards,

[yanqi liu]

yes，sir，just send data to me

Answer 2

[Jahetbe]

Dear @KSSV

Thank you for your response.

I want to use CNN to solve my problem.

My data is as follows.

Inputs = [ x11 x12 x13 x14

x21 x22 x23 x24

. . . .

xN1 xN2 XN3 XN4]

Outputs = [ Y11

Y21

.

.

.

.

YN1 ]

Wher N is the number of samples (i.e., 78000)