custom regression deep learning

Hi~

I ran into an error while doing custom regression.

In short, it is a neural network that receives 8 features as input and outputs 1 output.

My code and error are as follows.

clear,clc,close all
data=readmatrix('train.csv');
inputs=data(:,1:8);
targets=data(:,9);
input2=transpose(inputs);
target2=transpose(targets);
inputs2=normalize(input2,2,'range');
layer = mseRegressionLayer('mse');

layers = [

    featureInputLayer(8,'Name','in')
    fullyConnectedLayer(1,'Name','fc2')
    

    ];

lgraph=layerGraph(layers);
dlnet=dlnetwork(lgraph);

iteration = 1;
averageGrad = [];
averageSqGrad = [];
learnRate = 0.005;
gradDecay = 0.75;
sqGradDecay = 0.95;

dlX=dlarray(inputs2);
for it=1:5000
    iteration = iteration + 1;
    [gradient,loss]=dlfeval(@modelGradients,dlnet,dlX,target2);
    [dlX,averageGrad,averageSqGrad] = adamupdate(dlX,gradient,averageGrad,averageSqGrad,iteration,learnRate,gradDecay,sqGradDecay);
    if it>=4500 & mod(it,10)==0
        disp(it);
    end
end

function [gradient,loss]=modelGradients(dlnet,dlx,t)
    out=forward(dlnet,dlx);
    gradient=dlgradient(loss,dlx);
    loss=mean((out-t).^2);
end



Error using dlfeval (line 43)
First input argument must be a formatted dlarray.

Error in untitled3 (line 31)
    [gradient,loss]=dlfeval(@modelGradients,dlnet,dlX,target2);

    
Thank you for reading my question, and I hope someone who has insight will write an answer.

Answer 1

[Iuliu Ardelean]

Hey, when you call dlX=dlarray(inputs2), you should specify which dimensions are Spatial/Batch/Channel etc.

e.g.

X = randn(3,5);
dlX = dlarray(X,'SC')

SC are space and channel in this case.

Read more here:

https://uk.mathworks.com/help/deeplearning/ref/dlarray.html#mw_e439e5de-77e9-4399-a2e8-2cc62d2ab575

[jaehong kim]

oh! I solve the problem Thank you for your answer

[jaehong kim]

Hi

I have additional questions, can you answer me?

Thank you

[Iuliu Ardelean]

Hi Yeah -- What is the issue?

[jaehong kim]

Hi Thank you Now multivariate regression(y1,y2,y3...y6) is in progress. Regression with one output is good, but regression with more output is not good. To put it simply, W and B's updates are incorrect. Thus, all values of y_pred are equal (---). This is link about it.

https://kr.mathworks.com/matlabcentral/answers/743277-custom-regression-multiple-output?s_tid=srchtitle

Thank you for reading it. Any help is appreciated and welcome.

[Iuliu Ardelean]

Hey I've read the question and am not sure why this happens. I will try finding what is wrong tomorrow

[jaehong kim]

yes!

Thank you all the time for your hard work.