15
votes

My knowledge of matlab is merely on a need to know basis, so this is probably an elementary question. Nevertheless here it comes:

I have got a file containing data (16-bit integers) stored in binary format. How do I read it into a vector /an array in matlab? How do I write this data to a file in matlab? Is there any smart tweak to increase the performance speed when reading/writing a huge amount of data (gigabytes)?

3

3 Answers

16
votes

As Bill the Lizard wrote you can use fread to load the data into a vector. I just want to expand a little on his answer.

Reading Data

>> fid=fopen('data.bin','rb') % opens the file for reading
>> A = fread(fid, count, 'int16') % reads _count_ elements and stores them in A.

The commands fopen and fread default to Little-endian[1] encoding for the integers. If your file is Big-endian encoded you will need to change the fread to

>> A = fread(fid, count, 'int16', 'ieee-be');

Also, if you want to read the whole file set

>> count=inf;

and if you want to read the data into matrix with n columns use

>> count=[n inf];

Writing Data

As for witting the data to a file. The command, fwrite, in Bill's answer will write to a binary file. If you want to write the data to a text file you can use dlmwrite

>> dlmwrite('data.csv',A,',');

References

[1] http://en.wikipedia.org/wiki/Endianness

Update

  1. The machine format (IE, ieee-be, ieee-le, vaxd etc.) of the binary data can be specified in either the fopen or the fread commands in Matlab. Details of the supported machine format can be found in Matlab's documentation of fopen.

  2. Scott French's comment to Bill's answer suggests reading the data into an int16 variable. To do this use

    >> A = int16(fread(fid,count,precision,machineFormat));
    

    where count is the size/shape of the data to be read, precision is the data format, and machineformat is the encoding of each byte.

  3. See commands fseek to move around the file. For example,

    >> fseek(fid,0,'bof');
    

    will rewind the file to the beginning where bof stands for beginning of file.

10
votes

Assuming you know how many values you have stored in the file, you can do something like this to read the data into an array.

fid = fopen('data.bin','rb')
A = fread(fid, count, 'int16')

To write data to a file do this:

fid = fopen('data.bin','w')
count = fwrite(fid, A, 'int16')

The fwrite function returns the number of elements (not bytes) written to the file.

As far as performance tuning goes, you can read data in chunks to only use as much as you need to process. This is the same in any language, and there's no way to speed it up that's specific to Matlab.

0
votes

I usually hate seeing links in a response, but this looks pretty close:

http://www.mathworks.com/support/tech-notes/1400/1403.html

As to the second part of performance tuning, it's been 6 years since I've used Matlab, so I don't know.

HTH