How to restore a 2-dimensional numpy.array from a bytestring?

Question

numpy.array has a handy .tostring() method which produces a compact representation of the array as a bytestring. But how do I restore the original array from the bytestring? numpy.fromstring() only produces a 1-dimensional array, and there is no numpy.array.fromstring(). Seems like I ought to be able to provide a string, a shape, and a type, and go, but I can't find the function.

score 11 · Accepted Answer · answered Aug 23 '11 at 20:01

11

>>> x
array([[ 0.   ,  0.125,  0.25 ],
       [ 0.375,  0.5  ,  0.625],
       [ 0.75 ,  0.875,  1.   ]])
>>> s = x.tostring()
>>> numpy.fromstring(s)
array([ 0.   ,  0.125,  0.25 ,  0.375,  0.5  ,  0.625,  0.75 ,  0.875,  1.   ])
>>> y = numpy.fromstring(s).reshape((3, 3))
>>> y
array([[ 0.   ,  0.125,  0.25 ],
       [ 0.375,  0.5  ,  0.625],
       [ 0.75 ,  0.875,  1.   ]])

answered Aug 23 '11 at 20:01

Mike Graham

73,987
14
101
130

Didn't know that was even possible. This is exactly what I was looking for. Thanks. – David Eyk Aug 23 '11 at 20:20

score 0 · Answer 2 · answered Mar 27 '19 at 19:10

An update to Mike Graham's answer:

numpy.fromstring is depreciated and should be replaced by numpy.frombuffer
in case of complex numbers dtype should be defined explicitly

So the above example would become:

>>> x = numpy.array([[1, 2j], [3j, 4]])
>>> x
array([[1.+0.j, 0.+2.j],
       [0.+3.j, 4.+0.j]])
>>> s = x.tostring()
>>> y = numpy.frombuffer(s, dtype=x.dtype).reshape(x.shape)
>>> y
array([[1.+0.j, 0.+2.j],
       [0.+3.j, 4.+0.j]])

michel-slm · Answer 3 · 2011-08-23T18:25:42.323

0

It does not seem to exist; you can easily write it yourself, though:

def numpy_2darray_fromstring(s, nrows=1, dtype=float):
  chunk_size = len(s)/nrows
  return numpy.array([ numpy.fromstring(s[i*chunk_size:(i+1)*chunk_size], dtype=dtype)
                       for i in xrange(nrows) ])

edited Aug 23 '11 at 18:25

answered Aug 23 '11 at 18:14

michel-slm

9,438
3
32
31

`ndim` is a little misleading--that's specifying the number of rows, not the number of dimensions, by my reading. But it seems to work! – David Eyk Aug 23 '11 at 18:20
ah, you're right; serves me right for testing it on a test case where nrows = ndim = 2 ! Fixing the answer – michel-slm Aug 23 '11 at 18:22
Using the `reshape` method is more direct, readable, robust, and efficient than this, and less errorprone. – Mike Graham Aug 23 '11 at 20:04
Agreed, much more robust than this. – David Eyk Aug 23 '11 at 20:18
Good call. Likely much faster too! – michel-slm Aug 23 '11 at 20:37

How to restore a 2-dimensional numpy.array from a bytestring?

3 Answers3

Linked