How to extract the bits of larger numeric Numpy data types

Question

Numpy has a library function, np.unpackbits, which will unpack a uint8 into a bit vector of length 8. Is there a correspondingly fast way to unpack larger numeric types? E.g. uint16 or uint32. I am working on a question that involves frequent translation between numbers, for array indexing, and their bit vector representations, and the bottleneck is our pack and unpack functions.

Answer 1

You can do this with view and unpackbits

Input:

unpackbits(arange(2, dtype=uint16).view(uint8))

Output:

[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0]

For a = arange(int(1e6), dtype=uint16) this is pretty fast at around 7 ms on my machine

%%timeit
unpackbits(a.view(uint8))

100 loops, best of 3: 7.03 ms per loop

As for endianness, you'll have to look at http://docs.scipy.org/doc/numpy/user/basics.byteswapping.html and apply the suggestions there depending on your needs.