Good practice in conditional test with categorical variable

So I defined a categorical variable and tested its value in a conditional statement using the "==" operator, as seen in the image. The program works OK, thus I'm guessing such syntax is acceptable. But I kept getting warnings like the one in the image in the MATLAB IDE. How can my code be changed so that the warning is suppressed, and what is the underlying explanation for what is and is not a good practice when dealing with categorical variables in a conditional test?

Answer 1

[the cyclist]

The reason you are getting this mlint warning isn't really about the categorical variables at all. It is because you are trying to make a vector comparison in the if statement.

'def' is a 1x3 character vector, so MATLAB is warning you that it is expecting an expression that will evaluate to a scalar in the if statement. It is akin to writing

if [2 3] == [5 7 11]

Now, it turns out that because MATLAB will do a conversion to compare your categorical variable to the character array

catvar = categorical({'def'},{'abc','def'});
if catvar == 'def'
    fprintf("got here\n")
end

got here

the syntax will work. (I don't know if it gives the result you intend, though.) In general, it's better not to rely on the conversion, but make it explicit. I don't know your broader context for wanting to use categorical variables overall, but this syntax would not give the warning.

catvar = categorical({'def'},{'abc','def'});
if catvar == categorical({'def'})
    fprintf("got here, without mlint warning\n")
end

got here, without mlint warning

I do have the feeling that your whole situation might be easier if you used string arrays instead of categorical, but like I said, I don't know your broad context.

[Xiaoxing Zhang]

Yes, this method does the trick, thanks. I just wonder why MATLAB introduced the kind of usage that would be frowned upon by mlint in their documentation in the first place. Then again, that's not my problem now. The categorical helps for vvvery large varibles with limited number of types, it saves memory comparing to string arrays and are more informative than, say, a int8 array.

[the cyclist]

To be clear, the syntax

B == 'red'

is not what is throwing the warning. it is the use of that expression in the if statement

if B == 'red'

[Xiaoxing Zhang]

Ah yes，I think I got your point, if I wrap all or any around the == statement it would be fine, so basically == for categoricals expects returning of an array. That does sounds in line with other MATLAB logics.

Thanks a lot!

[Peter Perkins]

All true, but probably the best thing is

B == "red"

Strings make your life easier than cellstrs!

[Xiaoxing Zhang]

Good point...

[the cyclist]

Between @Peter Perkins and me, we probably have nearly 60 years of MATLAB experience, so if we both suggested using strings arrays, you know it's a good idea. :-)