Is it possible to increase the speed of writing a delimited text file using "writetable" and if so how?

Writing a table data structure containing 50,000 lines and 18 columns (all entries are numbers) takes about 50 seconds but reading the resulting file in with "readtable" takes about 0.8 seconds. I don't understand why writetable is so much slower than readtable.

I need to share the resulting file with people who don't use Matlab, so saving the table data structure as a *.mat binary file will not work.

Answer 1

[per isakson]

Caveat: I've never used the new &nbsp table !

• "I don't understand why ..." &nbsp I assume that regarding writetable speed was not a priority.
• "people who don't use Matlab" &nbsp I assume they want a text file.
• Timing "I/O" is a bit tricky because the result depends on the state of the system cache, which changes over time.
• fprintf &nbsp is the best way to write to text when it comes to speed.

I made a little experiment on R2013a and a 8 year old vanilla PC desktop. Run it on your computer. It indicates that a ten time increase in writing speed is possible.

>> et = cssm([50000,18]);
>> et
et =
    4.6681    0.4383    0.5839

where

function    et = cssm( sz ) 
%{
cssm([3,5]);
et = cssm([50000,18]);    
%}
    M1  = randi( 9, sz );
    tic
    frm = repmat( '%g ', [ 1, sz(2) ] );
    frm = [ frm(1:end-1), '\n' ];
    fid = fopen('c:\tmp\test.txt','w');
    for rr = 1 : sz(1)
        fprintf( fid, frm, M1( rr, : ) ); 
    end
    fclose( fid );
    et = toc;
    tic
    fid = fopen('c:\tmp\test.txt','r');
    M2  = textscan( fid, repmat('%f',[1,sz(2)])             ...
                ,   'Delimiter',' ', 'CollectOutput', true  );
    fclose( fid );
    et  = [ et, toc ];
    tic
    fid = fopen('c:\tmp\test.txt','r');
    M3  = fscanf( fid, '%f', sz );
    fclose( fid );
    et  = [ et, toc ];    
end

Thus try

array = table2array( your_numerical_table );
...
for rr = 1 : sz(1)
     fprintf( fid, frm, array( rr, : ) ); 
end