Question

I have two variables, x and y. x is

type(x) = <class 'numpy.matrixlib.defmatrix.matrix'>

type(y) = <type 'numpy.ndarray'>

x.shape = (869250, 1)

y.shape = (869250,)

x+y gives a MemoryError, despite the fact that I have around 5 gb free. This seems rather odd – does anyone have an idea as to what might be going on?

This is numpy 1.5.1, python 2.7, on 64 bit Linux.

Answer 1

Are you sure you’re doing what you want to be doing?

In [2]: x = np.random.normal(size=(500,1))

In [3]: y = np.random.normal(size=(500,))

In [4]: (x + y).shape
Out[4]: (500, 500)

This is a somewhat unintuitive application of numpy’s broadcasting rules. Your result is actually going to be 869250 x 869250, for a total of 5.5 terabytes of storage in the probably-default np.float64.

You more likely to want the vector sum. If you want to keep x as a matrix (which is often confusing, but…), you could do something like x + y.reshape(-1, 1).