Debugging inconsistent matrix indexing

I was wondering if you could answer my debugging-related question:

Suppose that my code includes the definition of a matrix, let's say:

X = randn(5,5) (a 2x2 matrix containing random numbers)

..but much later in my code, I accidentally forget that X is a matrix and I write another line of code where I treat it as a vector:

Y = X(5)

This does not cause a syntax error. And I understand why it's not supposed to. The question is one of debugging: Is there a shortcut that would help me easily detect, or warn me, if I accidentally treated some matrix as a vector (or, say, treated a 3-dimensional matrix as a 2-dimensional one, for example)?

I realize that, as a possible solution, I can keep track of my matrix sizes "by hand", and I can just "pay more attention", but I was wondering if there is a specific debugging method for the issue I describe?

Best regards,

Answer 1

[Jan]

Matlab allows to handle an array with different numbers of dimensions on purpose. The so called linear indexing is a typical example: You can index an array very efficiently using a single index:

x = rand(2,3);
disp(x(1:6));

Because this is not a bug, but intended behavior, there ist no standard "debugging" method to detect this automatically.

If this method means a bug in your code, using a user defined object would be better: Then you can define manually that a run-time error appears, when the object is indexed with the wrong number of dimension.

[Josu Uusitalo]

Ok, sure. Thanks for the quick answer!

I do need indexing like X(:) every now and then, if X is a matrix. Still it's interesting that I rarely, if ever, purposely use something like "X(5,5)" if X is actually 3-dimensional. Of course I would find use for X(5,5,:) or X(5,5,1), etc...

It could be nice if there was a debugging option, which (when turned on), X(5,5) in the above example would generate a note in the spirit of " did you really mean this?"

But I understand if this idea is not easy to implement (e.g. Matlab would have to make judgements about which types of syntaxes are "permitted but rather unusual").

Josu

[Stephen23]

The concept could equally be viewed from the other direction: "did you really mean to use subscript indexing?" or "did you really mean to use logical indexing?". There is nothing inherently better or correct about any one of these different indexing methods: they simply allow for different work flow, and make different tasks easier. That is why MATLAB supports them.

I use linear indexing quite a lot in my code while accessing elements of matrices. This is a very useful feature. Without linear indexing my life would be more difficult!

[Thorsten]

I agree with Stephen, but if someone like Josu does not uses this form of indexing, it would be nice if one could simple turn on a warning if less then ndims indices are used to adress a matrix of ndims >= 2.

[Josu Uusitalo]

Thanks Stephen and Thorsten,

I do use various types of indexing. But I haven't used something like X(5,5) if the actual size of X is, for example: 5,5,5.

If I enter X = randn(5,5,5), for example, and then if I use X(5,5), Matlab gives the same as when I type X(5,5,1).

I'm sure there's some logic behind this...but personally I might "enjoy" getting an error message in this situation...

But I understand the way 2-dimensional matrices can be addressed simply with one index. And that it can be useful. For example, if X = randn(5,5); then X(6) will give a valid result. It's not an error.

Personally I rarely use that, though. I would just use the two indices. Or X(5,:) to extract a vector from the matrix. Or sometimes I would use X(:) if I want to treat the matrix momentarily as just one "bunch of data".

[Jan]

Example of a useful purpose of linear indexing: Accesss data on the diagonal of a matrix:

A = rand(5, 5);
A(1:6:25) = 1;

Now the diagonal contains ones and the rest is random. Of course sub2ind would work also.

[Josu Uusitalo]

Thanks Jan, yeah your code does look somewhat nicer than for example:

(ones(5,5)-eye(5,5)).*randn(5,5) + ones(5,5)

..which came to my mind as I saw your comment.

Thanks!