Using np.savetxt and np.loadtxt with multi-dimensional arrays

Question

What is a generalized way of storing a more-than-2 dimensional array (ndim > 2) to file and retrieving it in same format (dimension) using np.savetxt and np.loadtxt?

My concern is if I give any delimiter while storing, do I need to give some treatments while retrieving? Plus dealing with floats and retrieving it in same format is little tricky.

I have seen many simple examples in the docs. I am just curious about whether the simplest storing np.savetxt(filename, array) can be retrieved using simply array = np.loadtxt(filename) or not.

Answer 1

If you need to save multi-dimensional arrays in a text file you can use the header parameter to save the original array shape:

import numpy as np

a = np.random.random((2, 3, 4, 5))

header = ','.join(map(str, a.shape))
np.savetxt('test.txt', a.reshape(-1, a.shape[-1]), header=header,
           delimiter=',')

And to load this array you can do:

with open('test.txt') as f:
    shape = map(int, f.next()[1:].split(','))
    b = np.genfromtxt(f, delimiter=',').reshape(shape)