Why does this code has an error of incompatible array size?

Question

Nazirul Mubin bin Afandi 2022-7-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1761305-why-does-this-code-has-an-error-of-incompatible-array-size

回答： Rohit 2023-8-31

%% load training data (use training_data.mat attached file)
% initialize number of nodes, weights, biases, errors and gradients.
%epochs and mini-batch size
clc
data = [out.dutycycle out.voltage out.current]
sample=1000;
labels = data(1:sample,1);
y = labels'; %output vector
images = data(1:sample,2:3);
images(:,1) = images(:,1)/400;
images(:,2) = images(:,2)/100;
images = images'; %Input vectors
hn1 = 80; %Number of neurons in the first hidden layer
hn2 = 60; %Number of neurons in the second hidden layer
%Initializing weights and biases
w12 = randn(hn1,1000).*sqrt(2/1000);
w23 = randn(hn2,hn1)*sqrt(2/hn1);
w34 = randn(1,hn2)*sqrt(2/hn2);
b12 = randn(hn1,1);
b23 = randn(hn2,1);
b34 = randn(1,1);
%learning rate
eta = 0.0058;
%Initializing errors and gradients
error4 = zeros(1,1);
error3 = zeros(hn2,1);
error2 = zeros(hn1,1);
errortot4 = zeros(1,1);
errortot3 = zeros(hn2,1);
errortot2 = zeros(hn1,1);
grad4 = zeros(1,1);
grad3 = zeros(hn2,1);
grad2 = zeros(hn1,1);
epochs = 50;
m = 10; %Minibatch size
%% Training phase 
for k = 1:epochs %Outer epoch loop
batches = 1;
for j = 1:sample/m
    error4 = zeros(1,1);
    error3 = zeros(hn2,1);
    error2 = zeros(hn1,1);
    errortot4 = zeros(1,1);
    errortot3 = zeros(hn2,1);
    errortot2 = zeros(hn1,1);
    grad4 = zeros(1,1);
    grad3 = zeros(hn2,1);
    grad2 = zeros(hn1,1);
for i = batches:batches+m-1 %Loop over each minibatch
%Feed forward
a1 = images(:,i);
z2 = w12.*a1 + b12;
a2 = elu(z2);
z3 = w23*a2 + b23;
a3 = elu(z3);
z4 = w34*a3 + b34;
a4 = elu(z4); %Output vector
%backpropagation
error4 = (a4-y(:,i)).*elup(z4);
error3 = (w34'*error4).*elup(z3);
error2 = (w23'*error3).*elup(z2);
errortot4 = errortot4 + error4;
errortot3 = errortot3 + error3;
errortot2 = errortot2 + error2;
grad4 = grad4 + error4*a3';
grad3 = grad3 + error3*a2';
grad2 = grad2 + error2*a1';
end
%Gradient descent
w34 = w34 - eta/m*grad4;
w23 = w23 - eta/m*grad3;
w12 = w12 - eta/m*grad2;
b34 = b34 - eta/m*errortot4;
b23 = b23 - eta/m*errortot3;
b12 = b12 - eta/m*errortot2;
batches = batches + m;
end
fprintf('Epochs:');
disp(k) %Track number of epochs
[images,y] = shuffle(images,y); %Shuffles order of the images for next epoch
end
disp('Training done!')
%note : create folder for this experiment and save the parameters.
% don't forget to keep the folder in matlab path!
save('wfour.mat','w34');
save('wthree.mat','w23');
save('wtwo.mat','w12');
save('bfour.mat','b34');
save('bthree.mat','b23');
save('btwo.mat','b12');
%% Testing phase 
% load testing data with labels ... (use testing_data.mat attached file)
testsample = 100;
labels = test(1:testsample,1);
y = labels';
images = test(1:testsample,2:3); 
images(:,1) = images(:,1)/400;
images(:,2) = images(:,2)/100;%copy
images = images';
we34 = matfile('wfour.mat');
w4 = we34.w34;
we23 = matfile('wthree.mat');
w3 = we23.w23;
we12 = matfile('wtwo.mat');
w2 = we12.w12;
bi34 = matfile('bfour.mat');
b4 = bi34.b34;
bi23 = matfile('bthree.mat');
b3 = bi23.b23;
bi12 = matfile('btwo.mat');
b2 = bi12.b12;
success = 0;
n = testsample;
for i = 1:n
out2 = elu(w2*images(:,i)+b2);
out3 = elu(w3*out2+b3);
out = elu(w4*out3+b4);
big = 0;
num = 0;
for k = 1:10
      if out(k) > big
         num = k-1;
         big = out(k);
end
end
if labels(i) == num
      success = success + 1;
end
end
fprintf('Accuracy: ');
fprintf('%f',success/n*100);
disp(' %');
%% ELU activation function
function fr = elu(x)
f = zeros(length(x),1);
for i = 1:length(x)
if x(i)>=0
    f(i) = x(i);
else
    f(i) = 0.2*(exp(x(i))-1);
end
end
fr = f;
end
%% derivative of the ELU activation function
function fr = elup(x)
f = zeros(length(x),1);
for i = 1:length(x)
if x(i)>=0
    f(i) = 1;
else
    f(i) = 0.2*exp(x(i));
end
end
fr = f;
end
%% SHuffle function 
function [B,v] = shuffle(A,y)
cols = size(A,2);
P = randperm(cols);
B = A(:,P);
v = y(:,P);
end
%Matlab Error: 
%Arrays have incompatible sizes for this operation.
% Error in ANN_PV_Array (line 57)
%     z2 = w12.*a1 + b12;

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

dpb 2022-7-16

Other than sticking the error at the end, this posting has the same shortcomings of your previous on the same subject.

Use the debugger and step through and you'll undoubtedly find the logic error -- we can't run it and it's far too much code with too few comments to be able to just scan it to find where you messed up on array sizes...

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Rohit 2023-8-31

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1761305-why-does-this-code-has-an-error-of-incompatible-array-size#answer_1298341

Hi Nazirul,

I understand that you are encountering an "incompatible array size" error in your code, it means that you are trying to perform a matrix operation with arrays of incompatible dimensions. This error commonly occurs when attempting operations like matrix multiplication or addition.

Here are some steps you can take to troubleshoot and resolve the "incompatible array size" error:

Verify the dimensions: Double-check the dimensions of the arrays involved in the operation. Ensure that the number of rows and columns match as expected. If necessary, print or display the dimensions of the arrays to verify their sizes.
Check the operation: Review the specific operation causing the error (e.g., matrix multiplication, element-wise multiplication, addition). Confirm that the operation is appropriate for the data and the desired result.
Transpose or reshape: If the dimensions are not compatible, consider transposing or reshaping the arrays to align their dimensions properly. Transposing changes the rows into columns and vice versa, while reshaping modifies the overall shape of the array.
Debug step by step: If the error persists, try debugging the code step by step. Print or display intermediate results to identify where the dimensions may have changed unexpectedly.

By carefully examining the dimensions of the arrays and verifying their compatibility, you should be able to resolve the "incompatible array size" error. If you encounter any specific error messages or need further assistance, please provide the relevant data files, and I will be glad to help you further.