New numpy array from attribute of objects in another numpy array

Question

Is there a way to do an element-wise attribute extraction from a numpy array? For example, say I have:

import numpy as np

class foo():
    def __init__(self, value):
        self.bar = value

obj_array = np.empty((2, 2), dtype='object')
for i in range(obj_array.shape[0]):
    for j in range(obj_array.shape[1]):
        obj_array[i, j] = foo(i+j)

bar_array_hard_way = np.empty_like(obj_array)
for i in range(obj_array.shape[0]):
    for j in range(obj_array.shape[1]):
        bar_array_hard_way[i, j] = obj_array[i, j].bar

Here, I have an array of objects. Each object has some attribute. I'm hoping there is a slick built-in way of extracting those attributes as a new numpy array. Obviously this is a pretty trivial example, but larger arrays the element-wise copying is pretty annoying.

This is almost certainly not working code, since obj_array.size[0] is invalid ... you mean obj_array.shape — donkopotamus
– donkopotamus, Commented Jun 8, 2016 at 23:06

jsbueno · Accepted Answer · 2016-06-08 23:41:23Z

I think the fastest way will be to combine Python's operator.attrgetter with numpy's np.frompyfunction - the first gives a fast, native code inlined, way to retrieve an object's attribute. The second, transforms an ordinary Python function into a Numpy's broadcast function, which can process an entire array in a single call -

so, your call is:

from operator import attrgetter
import numpy as np
# code to build obj_array
...
bar_array_easy_way = np.frompyfunc(attrgetter("bar"), 1, 1)(obj_array)

Quickly comparing it against using fromtiterator built a 1 million int array from my objects in half the time - besides, fromiterator can't build arrays with dtype=object - just fixed size elements.

Note that attrgetter itself is rather a "function factory" - it takes an attribute name, and returns a function that will take any object and return that attribute. That returned function we pass in turn, to frompyfunc - which takes other 2 arguments in order to allow numpy to make its broadcasting magic: the number of input arguments and the number of return results for our function.

donkopotamus · Accepted Answer · 2016-06-08 23:25:35Z

0

You can create your new array as:

bar_array = np.reshape(
    np.fromiter((x.bar for x in obj_array.flat), dtype=float),
    obj_array.shape)

Alter the dtype to whatever you'd please.

answered Jun 8, 2016 at 23:25

donkopotamus

23.4k3 gold badges58 silver badges61 bronze badges

Collectives™ on Stack Overflow

New numpy array from attribute of objects in another numpy array

2 Answers 2

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related