How to replace only part of a numpy array's values with those of another array?

Question

I'm trying to do something along the lines of a masked broadcast, where only certain values are broadcasted.

Suppose I have one bigger array, bigger_array, and one smaller array, smaller_array:

import numpy as np
import numpy.ma as ma

bigger_array = np.zeros((4,4), dtype=np.int32)
smaller_array = np.ones((2,2), dtype=np.int32)

Now, I only want the first three values of the smaller array to replace those of a certain section of the bigger array, but masking doesn't do what I'd hoped it would do:

masked_smaller_array = ma.masked_array(smaller_array, mask=[(0, 0), (0, 1)])
bigger_array[2:4, 2:4] = masked_smaller_array

This just returns the same thing a regular broadcast would, namely:

[[0 0 0 0]
 [0 0 0 0]
 [0 0 1 1]
 [0 0 1 1]]

Instead of my hoped for

[[0 0 0 0]
 [0 0 0 0]
 [0 0 1 1]
 [0 0 1 0]]

Stripping out the masked value before overriding via

bigger_array[2:4, 2:4] = masked_smaller_array[~masked_smaller_array.mask]

is also of no use as that flattens the array making broadcasting incompatible.

Is there some other way to achieve the same effect perhaps?

Divakar · Accepted Answer · 2015-12-19 21:53:38Z

You were close when you had -

bigger_array[2:4, 2:4] = masked_smaller_array[~masked_smaller_array.mask]

You were just needed to index the left side of the equation to use the same mask over there as well. Thus, one way to solve it would be to fix it, like so -

# Mask corresponding to smaller array from where elements are to be taken
select_mask = ~masked_smaller_array.mask

# Use the mask on source (smaller array) to select specific elements from it 
# and update sliced and masked (with same mask) places in bigger array
bigger_array[2:4, 2:4][select_mask] = smaller_array[select_mask]

Sample run -

In [59]: bigger_array = np.zeros((4,4), dtype=np.int32)
    ...: smaller_array = np.ones((2,2), dtype=np.int32)
    ...: masked_smaller_array =ma.masked_array(smaller_array,mask=[(0, 0),(0, 1)])
    ...: 

In [60]: select_mask = ~masked_smaller_array.mask

In [61]: select_mask
Out[61]: 
array([[ True,  True],
       [ True, False]], dtype=bool)

In [62]: bigger_array[2:4, 2:4][select_mask] = smaller_array[select_mask]

In [63]: bigger_array
Out[63]: 
array([[0, 0, 0, 0],
       [0, 0, 0, 0],
       [0, 0, 1, 1],
       [0, 0, 1, 0]], dtype=int32)

Anton Protopopov · Accepted Answer · 2015-12-19 21:55:27Z

0

Actually could use numpy.where with your mask as condition and fill it with smaller_array when condition is true. Btw IIUC you could use make_mask instead of masked_array for that purpose:

your_mask = [(0, 0), (0, 1)]
mask = ma.make_mask(your_mask)
np.where(~mask, smaller_array, bigger_array[2:4, 2:4])


In [106]: mask
Out[106]: 
array([[False, False],
       [False,  True]], dtype=bool)

In [108]: np.where(~mask, smaller_array, bigger_array[2:4, 2:4])
Out[108]: 
array([[1, 1],
       [1, 0]], dtype=int32)

Then you could assign that for your variable

answered Dec 19, 2015 at 21:55

Anton Protopopov

31.9k13 gold badges93 silver badges96 bronze badges

Collectives™ on Stack Overflow

How to replace only part of a numpy array's values with those of another array?

2 Answers 2

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related