Given a numpy array, is there a script that can be written, purely in python, that returns its compression ratio?
This is a very simple, specific problem that I can't seem to come up with a good solution for without manually making use of the file system.
Note that making use of the compressed file itself is irrelevant to this problem. The answer need only pertain to the compression ratio value.
As numPy uses c, I don't think a pure python solution is possible but you can avoid the file system using stringIO. Using numpy built in functions np.savez_compression we can then compare the resulting sizes to np.savez,
import StringIO
def get_compression_ratio(a):
uncompressed = StringIO.StringIO()
compressed = StringIO.StringIO()
np.savez_compressed(compressed, a)
np.savez(uncompressed, a)
return uncompressed.len/float(compressed.len)
a = np.zeros([1000,1000])
a[23,60] = 1.
b = np.random.random([1000,1000])
print("one number = ", get_compression_ratio(a),
"random = ", get_compression_ratio(b))
with result,
('one number = ', 1001.0255255255255, 'random = ', 1.0604228730260878)
As the random numbers are incompressible, this makes some sense but the array with one non-zero value should be better. The result relies on the algorithm in savez_compression being efficient/correct.
a = np.array([1, 2, 0, 3, 4, 0, 5]). Does it even make sense to ask for the "compression ratio" ofa? If you have something more specific in mind, please update the question.np.array([1,1,1,1,1,1,1])is more compressible thannp.array([1, 2, 0, 3, 4, 0, 5])for example.