CODEWITHSUNDEEP
pythonnumpy
Sean
Sean

Reputation: 4525

numpy element values do not change using indexing

a = [3,4,5,6]
b = [1,2,3]
adj = np.random.rand(10,10)
adj[a,:][:,b] = adj[a,:][:,b]  + 1000

Why do the element values of adj not change after adj[a,:][:,b] = adj[a,:][:,b] + 1000?

Upvotes: 2

Views: 79

Answers (2)

Jaime
Jaime

Reputation: 67437

As has been pointed out, fancy indexing always returns a copy, not a slice. So you are modifying a copy that is later discarded.

When indexing several dimensions with arrays, this get broadcasted to a common shape, so any of the following will also do the trick for you:

a = [[3], [4], [5], [6]]
b = [1, 2, 3]
adj[a, b] += 1000

a = np.array([3, 4, 5, 6]
b = [1, 2, 3]
adj[a[:, None], b] += 1000

And of course, what should be your first option for your actual indices, although it will not work if they are not all consecutive integers:

adj[3:7, 1:4] += 1000

Upvotes: 2

ebarr
ebarr

Reputation: 7842

So I suspect numpy can't return a view to that slice, so you are trying to modify a copy that gets garbage collected.

A solution is to use meshgrid. For example:

a = [3,4,5,6]
b = [1,2,3] 
adj = np.arange(7*7).reshape(7,7)
X,Y = np.meshgrid(a,b)
x,y = X.ravel(),Y.ravel()
adj[(x,y)]+=1000

Where adj is now:

array([[   0,    1,    2,    3,    4,    5,    6],
       [   7,    8,    9,   10,   11,   12,   13],
       [  14,   15,   16,   17,   18,   19,   20],
       [  21, 1022, 1023, 1024,   25,   26,   27],
       [  28, 1029, 1030, 1031,   32,   33,   34],
       [  35, 1036, 1037, 1038,   39,   40,   41],
       [  42, 1043, 1044, 1045,   46,   47,   48]])

Upvotes: 2

Related Questions