Btukfyl

Suppose I have two dictionaries with the same keys and all the values are 2x2 numpy arrays. Assumptions:

• dictionaries have the same keys


• every value is a 2x2 numpy array for all dictionaries and keys.

x1 and x2 are sample dictionaries.

x1 =  {k: np.random.randint(20, size=(2, 2)) for k in range(5)}

x2 =  {k: np.random.randint(20, size=(2, 2)) for k in range(5)}

I would like to add both x1 and x2 together by their keys and the result would be a new dictionary.

So if...

  x1[0] = [[1,2],[3,4]] 

and...

  x2[0] = [[10,20],[30,40]]

a new dictionary value when key = 0, would be...

  x_total[0] = [[11,22],[33,44]]

The next step would be to do this for many dictionaries with this structure. I was thinking of doing this in a for loop, but If there are more efficient solutions, I'd love to learn about them.

I've tried the approach below using the collections library

from collections import Counter

a = Counter(x1[0])

b = Counter(x2[0])

c = dict(a + b)

but I think this may not apply if the values are arrays.

I also know that np.add(x1[0], x2[0]) will result in the addition of the arrays but I'd like to do it across all keys at once.. if possible.

Simply use a dictionary comprehension:

{k: x1.get(k,0) + x2.get(k,0) for k in set(x1)}

For example:

import numpy as np



np.random.seed(0)



x1 =  {k: np.random.randint(20, size=(2, 2)) for k in range(5)}

x2 =  {k: np.random.randint(20, size=(2, 2)) for k in range(5)}

Yields:

{0: array([[12, 15],

       [ 0,  3]]), 1: array([[ 3,  7],

       [ 9, 19]]), 2: array([[18,  4],

       [ 6, 12]]), 3: array([[ 1,  6],

       [ 7, 14]]), 4: array([[17,  5],

       [13,  8]])}



{0: array([[ 9, 19],

       [16, 19]]), 1: array([[ 5, 15],

       [15,  0]]), 2: array([[18,  3],

       [17, 19]]), 3: array([[19, 19],

       [14,  7]]), 4: array([[0, 1],

       [9, 0]])}

Then applying our solution, we get:

{0: array([[21, 34],

       [16, 22]]), 1: array([[ 8, 22],

       [24, 19]]), 2: array([[36,  7],

       [23, 31]]), 3: array([[20, 25],

       [21, 21]]), 4: array([[17,  6],

       [22,  8]])}

Assuming all dictionaries are complete (they all have the same keys) a dict comprehension should be an efficient solution:

 x3 = {key: sum(x1[key] + x2[key]) for key in x1}

I'm using shorter datastructures for readability and assume that x1 and x2 share the same keys.

>>> x1 = {k:np.arange(k, k+2) for k in range(2)}

>>> x2 = {k:np.arange(k+1, k+3) for k in range(2)}

>>> 

>>> x1

{0: array([0, 1]), 1: array([1, 2])}

>>> x2

{0: array([1, 2]), 1: array([2, 3])}

>>>

>>> x_total = {k: x1[k] + x2[k] for k in x1}

>>> x_total

{0: array([1, 3]), 1: array([3, 5])}

Note that dictionaries with consecutive integer keys, especially when they start from zero, are a waste of memory (sparseness) and time (hashing). Why not use an array into which you can index with integers much more efficiently?