Vectorization: incompatible sizes of arrays

O is a matrix with n_p x n_o rows; P is a matrix with n_p rows. Both matrices have three columns.The second column in O should be filled with the average of two values in the second column of P: the first value at the row ip and the second value at randomly selected row.

counter = 0;

for ip = 1:n_p

for io = 1:n_o

counter = counter + 1;

O(counter,2) = (P(ip,2) + P(randi(n_p),2))/2;

end

end

I attempted to vectorize the code as:

ip = 1:n_p;

io = 1:n_o;

O(ip(:)+io(:),2) = (P(ip(:),2)+P(randi(n_p),2))/2;

I am receiving the following message: „Arrays have incompatible sizes for this operation“. Thank you for any suggestions.

Answer 1

[Torsten]

n_p = 3;
n_o = 3;
P = rand(n_p,2);
R = randi(n_p,n_p*n_o,1)

R = 9×1

3 2 2 2 3 1 3 2 1

IP = (cell2mat(arrayfun(@(i)vertcat(repmat(i,1,n_o)),1:n_p,'UniformOutput',0))).'

IP = 9×1

1 1 1 2 2 2 3 3 3

O(:,2) = (P(IP,2)+P(R,2))/2

O = 9×2

0 0.8950 0 0.5663 0 0.5663 0 0.2625 0 0.5912 0 0.5663 0 0.9199 0 0.5912 0 0.8950

[Milan Lstiburek]

Thank you. I like this solution and I learned something new.

[Bruno Luong]

IP = (cell2mat(arrayfun(@(i)vertcat(repmat(i,1,n_o)),1:n_p,'UniformOutput',0))).'

can be simplified to

IP = replem((1:n_p)',n_o)

[Torsten]

Or:

IP = reshape(repmat(1:n_p,n_o,1),n_p*n_o,1)

But yours is shorter :-)

Answer 2

[Bruno Luong]

Simpler one liner

O(:,2) = P(randi(end,n_o,end),2) + repelem(P(:,2),n_o) / 2;

Answer 3

[Bruno Luong]

O(:,2) = reshape(reshape(P(randi(end,n_o,end),2),n_o,[]) + reshape(P(:,2),1,[]), [],1) / 2;

[Milan Lstiburek]

I just checked and it seems to work equally. I like your approach very much, never thought this could be done in a single line. Thank you!

[Bruno Luong]

Good.

It is not easy to check the correctness due to randomness.