How is this even logical?

As you can see, y1 is a column vector.

When inserted at position 1 in y, it becomes a row vector.

Why and how do I not make it do this?

[Stephen23]

"When inserted at position 1 in y, it becomes a row vector."

Your description is incorrect: the indexing y(1,:) is not "position 1", it is the entire first row of y. Not the same thing at all.

In any case, you did not explain what you expect to occur: what should MATLAB do when you try to force three rows of data into one row?

Answer 1

[James Tursa]

When assigning into rows or columns via indexing, MATLAB will conform a RHS vector from row or column to row or column if necessary for the assignment.

E.g.,

x = 1:3 % a row vector

x = 1×3

1 2 3

z = zeros(3)

z = 3×3

0 0 0 0 0 0 0 0 0

z(1,:) = x % assign elements of x into 1st row, works as expected

z = 3×3

1 2 3 0 0 0 0 0 0

z(:,1) = x % this works too because of the auto conforming behavior, into a column

z = 3×3

1 2 3 2 0 0 3 0 0

z(1:4:end) = x % and assigning into linear indexing also works into the diagonal

z = 3×3

1 2 3 2 2 0 3 0 3

If you don't want this behavior, please provide an example of all inputs and desired output and we can probably help you with the syntax.

Answer 2

[Walter Roberson]

When inserted at position 1 in y

However, assigning to y(1,:) is not asking to insert at "position 1" -- the vector is length 3 and "position 1" would be a scalar, and you cannot store a vector with three elements into a location only big enough for a scalar.

Assigning to y(1,:) is asking to assign to row 1.

With the value to be assigned being a column vector, there are only two reasonable things that MATLAB could do:

1. Give an error message because the shape of the destination is not exactly the shape of the source; or
2. Go ahead and silently reshape the source column to fit into the destination row.

Giving an error message would certainly be a possibility. But is it useful ?

Suppose I gave you a task: Initialize A to be a 4 x 1 array of zeros, and copy the first three elements of the input vector y into the first three elements of A.

With the way MATLAB works at the moment, that becomes very easy:

A = zeros(4,1); A(1:3) = y(1:3);

But if MATLAB gave error messages for assigning between rows and columns (of the same number of elements) then it would have to look more like

A = zeros(4,1);
if size(y,1) == 1;
    A(1:3) = y(1:3).';
else
    A(1:3) = y(1:3);
end

or

A = zeros(4,1); A(1:3) = reshape(y(1:3),[],1);

Now suppose the task is to accept an input vector A and copy the first three elements of y into the first three elements of A.

if size(A,1) == 1;
    A(1,1:3) = reshape(y(1:3), 1, []);
else
    A(1:3,1) = reshape(y(1:3), [], 1);
end

Yes, it is do-able, but some kind of "type purity" of "row vector is a different type than a column vector" truly important compared to the convenience of being able to just do A(1:3) = y(1:3); ?