Numpy array to gzip file

[Robert Kern]

Sean Davis wrote:
> I have a set of numpy arrays which I would like to save to a gzip
> file.  Here is an example without gzip:
> 
> b=numpy.ones(1000000,dtype=numpy.uint8)
> a=numpy.zeros(1000000,dtype=numpy.uint8)
> fd = file('test.dat','wb')
> a.tofile(fd)
> b.tofile(fd)
> fd.close()
> 
> This works fine.  However, this does not:
> 
> fd = gzip.open('test.dat','wb')
> a.tofile(fd)
> 
> Traceback (most recent call last):
>   File "<stdin>", line 1, in <module>
> IOError: first argument must be a string or open file

As drobinow says, the .tofile() method needs an actual file object with a real 
FILE* pointer underneath it. You will need to call fd.write() on strings (or 
buffers) made from the arrays instead. If your arrays are large (as they must be 
if compression helps), then you will probably want to split it up. Use 
numpy.array_split() to do this. For example:

In [13]: import numpy

In [14]: a=numpy.zeros(1000000,dtype=numpy.uint8)

In [15]: chunk_size = 256*1024

In [17]: import gzip

In [18]: fd = gzip.open('foo.gz', 'wb')

In [19]: for chunk in numpy.array_split(a, len(a) // chunk_size):
    ....:     fd.write(buffer(chunk))
    ....:

> In the bigger picture, I want to be able to write multiple numpy
> arrays with some metadata to a binary file for very fast reading, and
> these arrays are pretty compressible (strings of small integers), so I
> can probably benefit in speed and file size by gzipping.

File size perhaps, but I suspect the speed gains you get will be swamped by the 
Python-level manipulation you will have to do to reconstruct the array. You will 
have to read in (partial!) strings and then put the data into an array. If you 
think compression will really help, look into PyTables. It uses the HDF5 library 
which includes the ability to compress arrays with gzip and other compression 
schemes. All of the decompression happens in C, so you don't have to do all of 
the manipulations at the Python level. If you stand to gain anything from 
compression, this is the best way to find out and probably the best way to 
implement it, too.

   http://www.pytables.org

If you have more numpy questions, you will probably want to ask on the numpy 
mailing list:

   http://www.scipy.org/Mailing_Lists

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
  that is made terrible by our own mad attempt to interpret it as though it had
  an underlying truth."
   -- Umberto Eco