sum parts of numpy.array

Question

Let's say I have the following array:

a = np.array([[1,2,3,4,5,6], 
              [7,8,9,10,11,12],
              [3,5,6,7,8,9]])

I want to sum the first two values of the first row: 1+2 = 3, then next two values: 3+4 = 7, and then 5+6 = 11, and so on for every row. My desired output is this:

array([[ 3,  7, 11],
       [15, 19, 23],
       [ 8, 13, 17]])

I have the following solution:

def sum_chunks(x, chunk_size):
    rows, cols = x.shape
    x = x.reshape(x.size / chunk_size, chunk_size)
    return x.sum(axis=1).reshape(rows, cols/chunk_size)

But it feels unnecessarily complicated, is there a better way? Perhaps a built-in?

nneonneo · Accepted Answer · 2013-09-03 00:57:50Z

8

Just use slicing:

a[:,::2] + a[:,1::2]

This takes the array formed by every even-indexed column (::2), and adds it to the array formed by every odd-indexed column (1::2).

answered Sep 3, 2013 at 0:57

nneonneo

181k37 gold badges331 silver badges412 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

Akavall Over a year ago

Thank you, slicing works nice for "2" case, but would it still be as nice for a general solution?

nneonneo Over a year ago

How would you generalize this? You can do np.sum([a[:,i::n] for i in xrange(n)], 0) if you might need to sum n consecutive columns, for example.

Akavall Over a year ago

Your solution is shorter and easier to read, but I timed it and my initial solutions is considerably faster.

nneonneo Over a year ago

@Akavall: the code in my answer is faster than your sum_chunks specifically for n=2. For larger n, other solutions may be faster. My comment solution is an example of a solution, but I was not aiming to optimize.

Jaime · Accepted Answer · 2013-09-03 02:33:51Z

When I have to do this kind of stuff, I prefer reshaping the 2D array into a 3D array, then collapse the extra dimension with np.sum. Generalizing it to n-dimensional arrays, you could do something like this:

def sum_chunk(x, chunk_size, axis=-1):
    shape = x.shape
    if axis < 0:
        axis += x.ndim
    shape = shape[:axis] + (-1, chunk_size) + shape[axis+1:]
    x = x.reshape(shape)
    return x.sum(axis=axis+1)

>>> a = np.arange(24).reshape(4, 6)
>>> a
array([[ 0,  1,  2,  3,  4,  5],
       [ 6,  7,  8,  9, 10, 11],
       [12, 13, 14, 15, 16, 17],
       [18, 19, 20, 21, 22, 23]])
>>> sum_chunk(a, 2)
array([[ 1,  5,  9],
       [13, 17, 21],
       [25, 29, 33],
       [37, 41, 45]])
>>> sum_chunk(a, 2, axis=0)
array([[ 6,  8, 10, 12, 14, 16],
       [30, 32, 34, 36, 38, 40]])

senderle · Accepted Answer · 2013-09-03 00:57:49Z

1

Here's one way:

>>> a[:,::2] + a[:,1::2]
array([[ 3,  7, 11],
       [15, 19, 23],
       [ 8, 13, 17]])

answered Sep 3, 2013 at 0:57

senderle

152k36 gold badges218 silver badges244 bronze badges

Collectives™ on Stack Overflow

sum parts of numpy.array

3 Answers 3

4 Comments

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

4 Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related