"char" doesn't always maintain orientation of input array

The "char" function maintains orientation of input arrays except for categorical arrays:

>> categorical(["A" "B"])
ans = 1×2 categorical array
     A      B 
>> char(categorical(["A" "B"]))
ans = 2×1 char array
    'A'
    'B'

Does anyone know the rationale behind this?

Answer 1

[Voss]

I imagine the rationale is that, in case the elements of the categorical array are multiple characters long, e.g.:

categorical(["Abbot" "Beto"])

ans = 1×2 categorical array
     Abbot      Beto 

You want to be able to know where one ends and the next one begins, so you put them on different rows:

char(categorical(["Abbot" "Beto"]))

ans = 2×5 char array
    'Abbot'
    'Beto '

as opposed to having the result be:

'AbbotBeto'

ans = 'AbbotBeto'

which can be ambiguous:

categorical(["0" "1" "10" "11"])

ans = 1×4 categorical array
     0      1      10      11 

char(categorical(["0" "1" "10" "11"]))

ans = 4×2 char array
    '0 '
    '1 '
    '10'
    '11'

since in this case '011011' could be interpreted as '0', '1', '10', '11', or as '0', '11', '0', '11', for example.

Just a guess.

[FM]

Thanks for the speculation, Voss. I see this as trading one kind of ambiguity for another. The new ambiguity is that you can't tell whether the original array was a row or column vector.

It's odd that char doesn't use the same approach for string and categorical arrays:

>> char(["dog" "catty"])

     1×5×2 char array
   ans(:,:,1) = 'dog  '
   ans(:,:,2) = 'catty'

>> char(categorical(["dog" "catty"]))

   ans = 2×5 char array
       'dog  '
       'catty'

In fact, the categorical array is treated the same as a cell array of char arrays:

>> char({'dog';'catty'})

   ans = 2×5 char array
       'dog  '
       'catty'