numpy.packbits doesn't accept a bool array

[Dunes]

Example of problem

data = numpy.array([True])
byte_values = numpy.packbits(data)

This results with an exception -- TypeError: Expected an input array of integer data type. This seems to go against the documentation of the function, which says it expects a 'binary-valued array'.

To make this work, you need to first convert the data type of the array to int, which seems redundant. eg.

byte_values = numpy.packbits(numpy.array(data, dtype=int))

A better use case of where I encountered the problem.

data = numpy.random.sample(1024)
truth_values = data >= threshold
bits = numpy.packbits(truth_values)

[LuisBL]

This is a real problem, because packbits is the only workaround the fact that np.array of boolean use 8bit by boolean. This is a limit to how compact a memory structure can be.

[LuisBL]

In [8]: x = np.array([0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 1], dtype=np.bool)

In [9]: px = np.packbits(x)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-9-326e19e2360b> in <module>()
----> 1 px = np.packbits(x)

TypeError: Expected an input array of integer data type
In [10]

[LuisBL]

There is a question on SO about this issue http://stackoverflow.com/questions/34511362/packing-numpy-boolean-array-needs-tons-of-ram

[alimuldal]

@LuisBL There is no need to cast the input array to the native int dtype - you can avoid generating a copy simply by viewing it as np.uint8, which also uses a single byte per element, e.g.:

px = np.packbits(x.view(np.uint8))

[jaimefrio]

That feature was added in #5319, it does work in current master, so closing this.