Using Numpy stride_tricks to get non-overlapping array blocks

Question

I\'m trying to using numpy.lib.stride_tricks.as_strided to iterate over non-overlapping blocks of an array, but I\'m having trouble finding documentation of the parameters,

Answer 1

Using @unutbu's answer as an example, I wrote a function that implements this tiling trick for any ND array. See below for link to source.

>>> a = numpy.arange(1,21).reshape(4,5)

>>> print a
[[ 1  2  3  4  5]
 [ 6  7  8  9 10]
 [11 12 13 14 15]
 [16 17 18 19 20]]

>>> blocks = blockwise_view(a, blockshape=(2,2), require_aligned_blocks=False)

>>> print blocks
[[[[ 1 2]
   [ 6 7]]

  [[ 3 4]
   [ 8 9]]]


 [[[11 12]
   [16 17]]

  [[13 14]
   [18 19]]]]

[blockwise_view.py] [test_blockwise_view.py]