difference between str2num and str2double for getting a matrix

>> txt = '5 7; 8 9';
>> f = str2num(txt)

f =

     5     7
     8     9

>> f = str2double(txt)

f =

   NaN
   

So I have a text of that form with semi-colon and spaces and the str2num() is doing a very good job in getting me a 2x2 matrix, which is what I need and it has saved me a lot of hassle.

But the software recommended to use str2double() as it has got high precission or something like that. So, my doubt is-

1) Why isn't it working like str2num() here?

2) What is the actual difference between them ?

3) Should I stick with str2num() in this case?

Answer 1

[Bruno Luong]

Under the hood

• str2num uses EVAL
• str2double uses SCANF, and work only on scalar.

I see no reason why SCANF is more accurate than EVAL. Both are limited by the digit number that is stored in string.

Now some people will tell you EVAL is evil. Honestly I don't care if it translates your string data directly to vector/matrix then it does the job and just use it.

Personally I use str2double if I know I want a scalar, it can be more rigorous error checking, all other case I use str2num.

I would be happy for someone to teach me otherwise.

[Steven Lord]

Just keep in mind the story of Bobby Tables if you're using str2num.

[Bruno Luong]

>> str = 'system(''echo "I remember BOB TABLE"'')';
>> x = str2num(str)
"I remember BOB TABLE" 

x =

     0

>>