row-column confusion with squeeze

Can anybody please explain why A has 3x1 dimension whereas B has 2x3 dimension?

I mean, in both cases, I put 3 as the last dimension in the 3D matrix. So, both A and B should have either 3 rows or 3 columns, right?

If the argument is, squeeze just get rid of all the single-tone dimensions without keeping the relative position, then A and C should be of same dimension? But that is not the case as well.

A=squeeze(zeros(1,1,3))

A =

       0
       0
       0

B=squeeze(zeros(1,2,3))

B =

       0     0     0
       0     0     0

C=squeeze(zeros(1,3,1))

C =

       0     0     0

[David Goodmanson]

but both A and B do have either 3 rows or 3 columns

[Abdullah Fahim]

Nope, A has 3 rows and B has 3 columns. I was expecting both A and B would have 3 rows OR both A and B would have 3 columns.

[KSSV]

Check this link for discussion on the same: https://in.mathworks.com/matlabcentral/answers/6560-how-does-squeeze-work

[Abdullah Fahim]

Thanks a lot Siva. I actually found that thread just after posting a new thread (sorry for that) and had commented on that discussion already. Just want to know the opinion from others, is the behavior logical?

[David Goodmanson]

Probably not real logical. It seems reasonable that (3,1,1,1) should become a column vector of size (3,1) (which happens) and (1,1,1,3) should become a row vector of size (1,3) (which doesn't happen). But then should ( [m ones] 3 [n ones] ) become a row vector or a column vector?

[Abdullah Fahim]

Good question.. I have no idea :). I got your point, so dropping the subject.

Answer 1

[Thorsten]

Squeeze squeezes out all singleton dimensions of X, if X is not a matrix.

In case of

C = squeeze(zeros(1,3,1))
whos C
Name      Size            Bytes  Class     Attributes

C         1x3                24  double

you already have a matrix, namely a 1 x 3 matrix. The trailing singleton dimensions are ignored by zeros.

isequal(zeros(1,3,1), zeros(1,3))
ans =
     1

or

isequal(zeros(1,3,1,1,1,1,1), zeros(1,3))
ans =
     1

So in case of C, nothing is squeezed and C is left as is.

In case of A, you don't have a matrix, but a 1 x 1 x 3 array. The first two singleton dimensions are squeezed out, resulting in 3 as the first dimension. Because there is no vector in Matlab, a second dimension is needed, which is 1. So A 1 x 1 x 3 is squeezed to 3 x 1.

In case of B, you have a 1 x 2 x 3 array. The first dimension is a singleton dimension which is squeezed out, resulting in a 2 x 3 matrix.

One could see an inconsistency between the handling of A and C, because the squeezing algorithm is not applied to C, since C is a matrix. But the help of squeeze explicitly states

2-D arrays are unaffected by squeeze so that row vectors remain rows.

BTW, squeeze is not built-in, so you can view the code with

edit squeeze

or

 type squeeze

and have a look at how it works.

[Abdullah Fahim]

Thanks Thorsten. I didn't realize zeros(1,3,1,1,1) === zeros(1,3). Thanks for pointing that out.