How do I import a file containing both numbers and strings

I've looked around and found multiple posts regarding this issue, however, none of them really helped me with this problem.

I'm trying to import a text file cointaining a square matrix, defined by the user, that contains values and the string X like, for example, this one:

17.01 24.02 1.03 8.04 
23.06 5.07 7.08 14.09 
4.01 X 13.03 20.04 
10.06 12.07 19.08 21.09 
11.01 18.02 25.03 2.04 

I've tried using importdata but when it reaches the X, the rest of the line gets replaced by NaN and the lines underneath get ignored. This is what happens:

17.0100   24.0200    1.0300    8.0400   
23.0600    5.0700    7.0800   14.0900   
4.0100       NaN       NaN       NaN      

Since the matrix is given by the user (randomly), its size and the 'X' positions will always be different. Knowing that the matrixes are always square, what can I do to solve this?

Answer 1

[Stephen23]

Use textscan and set its 'TreatAsEmpty' option to 'X'.

This will be much more efficient than importing and post-processing (e.g. with regexp).

[TADA]

it is about 2-3 times faster than my solution

but how do you solve the part about not knowing the size of the matrix?

granted the matrix is square you can always reshape to the square root of the length

but if it isn't, do you have a solution using textscan?

this is how far i got:

fid = fopen('blah.dat');
x1 = textscan(fid, '%n', 'TreatAsEmpty', 'X');
a = x1{:}; 

% this works well with the example
% but with a different size non-square matrix, it probably won't
A1 = reshape(a, floor(sqrt(length(a))), ceil(sqrt(length(a))))';

By the way, this is not me arguing, but trying to learn... :)

[Stephen23]

"but how do you solve the part about not knowing the size of the matrix?"

Read the first line using fgetl, count the delimiters, then frewind back to start of the file. Use the count to define a format string using repmat, then read the file using textscan. It sounds complex, but it isn't really:

opt = {'Delimiter',' ','TreatAsEmpty','X','CollectOutput',true};
[fid,msg] = fopen('temp1.txt','rt');
assert(fid>=3,msg)
cnt = nnz(strtrim(fgetl(fid))==' ');
frewind(fid)
fmt = repmat('%f',1,1+cnt);
C = textscan(fid,fmt,opt{:});
fclose(fid);

And checking (the test file is attached):

>> C{1}
ans =
   17.0100   24.0200    1.0300    8.0400
   23.0600    5.0700    7.0800   14.0900
    4.0100       NaN   13.0300   20.0400
   10.0600   12.0700   19.0800   21.0900
   11.0100   18.0200   25.0300    2.0400

[TADA]

nice +1

still much faster

Answer 2

[TADA]

txt = fileread('blah.dat');
lines = strsplit(txt, newline);
x = regexp(lines, '[^ ]+', 'match');
items = cat(1,x{:});
A = str2double(items);

your X would now be represented by NaN in matrix A

Answer 3

[dpb]

If you're going to read in text in random locations, unless you can determine or require the user to tell you where those locations are, the only alternative will be to read the whole array as cellstr array and then figure out after the fact "who's who in the zoo" as far as which locations are/aren't numeric.