Read one column of unknown length from CSV file

Let's say I want to read column C from a CSV file which has an unknown number of rows. I have written the following code:

csvread('Filepath',0,2,[0 2 Inf 2])

Which gives me the error:

Error using dlmread (line 157)
Internal size mismatch

Error in csvread (line 50)
    m=dlmread(filename, ',', r, c, rng);

Have also tried:

csvread('Filepath',0,2,[0 2 : 2])

Which gives the error:

Attempted to access range(3); index out of bounds because numel(range)=2.

Error in dlmread (line 105)
    if r > range(3) || c > range(4), result= []; return, end

Error in csvread (line 50)
    m=dlmread(filename, ',', r, c, rng);

So any suggestions on how to accomplish this? Thanks

Answer 1

[per isakson]

I failed. Since csvread is based on textscan, I propose you use textscan. It's at least better documented - IMO.

These two scripts read the third column of csv_test.txt.

fid = fopen( 'csv_test.txt', 'r' ); 
cac = textscan( fid, '%*d%*d%d%*d%*d%*d', 'Delimiter',',', 'Headerlines',1 );
fclose( fid );
fid = fopen( 'csv_test.txt', 'r' ); 
cac = textscan( fid, '%d%*[^\n]', 'Delimiter',','  ... 
             , 'Headerlines',1, 'Headercolumns',2 );
fclose( fid );

( 'Headercolumns',2 &nbsp is an undocumented option I stumbled upon when I tried to find out how to use csvread.)

and where csv_test.txt contains

A,B,C,D,E,F 
11,12,13,14,15,16
21,22,23,24,25,26
31,32,33,34,35,36

[Jacee Johnson]

Thanks to the both of you for your help. This seems like the best approach but now I have trouble using fopen for multiple files. I will create another thread for that topic.

Answer 2

[Walter Roberson]

csvread() invokes dlmread()

dlmread() invokes textscan() with the format set to '' (the empty string), which invokes an undocumented mode of textscan that allows it to try to determine the number of columns that are present. It also passes in the undocumented 'headercolumns' parameter, which is 0 to skip no columns and is otherwise the starting column to read from. It also uses the undocumented feature of passing -1 as the number of rows to textscan() instead of using the documented inf to mean infinite number.

The memory inconsistency error is disabled in dlmread if the number of rows is negative. And here is a hack: the number of rows will be -1 (telling textscan to read the entire file) if the third element of the range you specify is precisely 2 less than the starting row number that you give. So... instead of using

csvread('Filepath',0,2,[0 2 Inf 2])

invoke

csvread('Filepath',0,2,[0 2 -2 2])

-2 minus 0 gives -2, 1 is added to that in the internal calculation, -1 would be the result, that gets passed to textscan() as the number of rows so it reads everything, then the fact that the number of rows is negative disables the mismatch error.

By the way, when you specify a column range to csvread() it passes it to dlmread(). When dlmread() is given a column range, it uses the undocumented headercolumns parameter to tell textscan() the starting column. textscan() reads all of the columns from there to the end of the line, and returns all those columns. And then dlmread() throws away any extra columns.

So all in all, you might as well invoke textscan() directly and use an '%*' specifier to tell textscan to not even return data for the columns you don't want, the way that per isakson showed.

[Jacee Johnson]

Thank you for your response and your method is interesting and would be a more efficient way of accomplishing my goal since there are a large number of columns and using the %* specifier for such a large number of columns is tedious.

I tried your suggestion and cannot seem to get it work, however. I used your exact code on a sample .CSV file I created and I cannot get it to read any values unfortunately. If I am forgetting something please don't hesitate to point it out.

[Image Analyst]

Jacee, you make it very hard for people to help you when you don't do things like attach your .CSV files so they can try the code on your actual data before suggesting the code to you.

[per isakson]

To avoid those error-prone format strings use repmat, e.g.

[ repmat('%*s',1,17),'%f', etc. ]

[Jacee Johnson]

Per, thanks once again, this command made my code much simpler and easier to manage. Thank you very much for all of your help.

[Nicholas Paul Patzke]

passing along the -1 to text scan made it work for me!!!!