Why does outputing numpy.dot to memmap does not work?

Question

If I do:

a = np.ones((10,1))
b = np.ones((10,1))
c = np.memmap('zeros.mat', dtype=np.float64, mode='w+', shape=(10,10), order='C')

a.dot(b.T, out=c)

I am getting:

ValueError: output array is not acceptable (must have the right type, nr dimensions, and be a C-Array)

I check all conditions from the error message and they seem to fit:

>>> print(a.dtype == b.dtype == c.dtype)
>>> print(np.dot(a, b.T).shape == c.shape)
>>> print(c.flags['C_CONTIGUOUS'])

True
True
True

When I replace c with:

c = np.zeros((10,10))

it works.

What am I doing wrong?

Answer 1

It doesn't just have to match the dtype; it also has to have the right type , as in type(c) . c is a numpy.memmap instance, not a numpy.ndarray , so that check fails.

As recommended in the numpy.memmap docs , you could instead use mmap.mmap to map the file and create a numpy.ndarray backed by the mmap as its buffer. You can look at the numpy.memmap implementation to see what might be involved in doing that.

Answer 2

From RKI's comment, use of numpy.asarray works directly, eg:

a = np.ones((10,1))
b = np.ones((10,1))

c_memmap = np.memmap('zeros.mat', dtype=np.float64, mode='w+', shape=(10,10), order='C')
c = numpy.asarray(c_memmap)

a.dot(b.T, out=c)

c_memmap.flush()
#etc.