How can I determine the number of Headerlines for varying, non-rectangular text files so that I can parse it with textscan?

I would like to use textscan to read in the tabular integer and floating point data, keying off of the *NODE line. This line can be anywhere in the file with other, non-comment strings and integer lines in there as well. How can I find the varying number of headerlines for any given input file?

My code and example input file are as follows, Thanks!

fid4 = fopen('E:\scratch\ANSYS_macro\MATLAB dyna beams\sample.k'); g = textscan(fid4,'%d %f %f %f','Delimiter','\n','headerlines',15); celldisp(g); fclose(fid4);

*KEYWORD
*TITLE

*DATABASE_FORMAT
$   1IFORM  2IBINARY
         0
$
$
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$                               NODE DEFINITIONS                               $
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$
*NODE
$   1NID              2X              3Y              4Z     5TC     6RC
       1        0.141746         0.55315     -0.00592088
       2        0.141746        0.538028     -0.00592088
       3        0.126928         0.55315     -0.00669746
       4        0.126926        0.538027     -0.00669757
       5        0.112141         0.55315     -0.00747244
       6        0.112138        0.538025     -0.00747256
       7       0.0973459         0.55315      -0.0082478
       8       0.0973435        0.538024     -0.00824792
       9       0.0825538         0.55315     -0.00902302
      10       0.0825514        0.538022     -0.00902315
      11       0.0677682         0.55315      -0.0097979
$

Answer 1

[per isakson]

I'm not sure I understand whether your file contains one or more blocks of numerical data. Here is a file that handles both cases.

    function    g = ccsm()
        str = fileread( 'cssm.txt' );
        cac = regexp(str,'(?<=\*NODE\s+).+?(?=((\*KEYWORD)|($)))','match');
        g   = cell( 1, length( cac ) ); 
        for jj = 1 : length( cac )
            g{jj} = textscan(cac{jj},'%d%f%f%f', 'CommentStyle','$');        
        end
    end

returns a cell array g, where

    >> g{3}
    ans = 
        [11x1 int32]    [11x1 double]    [11x1 double]    [11x1 double]

and where cssm.txt contains three copies of text you included in your question.

.

Comments:

• the entire text file must fit in memory to use this approach
• it is not possible read and parse the file in one step with textscan
• it is safer to use a definition of the file format than guess based on one sample
• regexp is powerful and fast, but ... . The expression I used assumes that blocks of numerical data are enclosed by "*NOTE" and "*KEYWORD" or by "*NOTE" and end of file.

[Cedric]

Ah, a regexp challenge, I take it! ;-)

I'd propose the following:

 blocks = regexp( content, '\*NODE(.*?\n){2}([\s\d\-\.]+)', 'tokens' ) ;

if the block doesn't always end with a $ character, and

 blocks = regexp( content, '\*NODE(.*?\n){2}([^$]+)', 'tokens' ) ;

if it does. Then, blocks{1} (only cell if there is only one block) is a cell array whose cell 1 contains the header, and whose cell 2 contains the data.