CODEWITHSUNDEEP
pythonarraysnumpy
AfonsoSalgadoSousa
AfonsoSalgadoSousa

Reputation: 405

Double Indexing Assignment in NumPy

I would like to know how can I assign a matrix to a slice of a slice of a NumPy array. Following I have toy data and a 1D mask to illustrates what I need to do:

data = np.array([[1, 2, 11, 21], [2, 4, 12, 23], [3, 6, 13, 25], [4, 8, 14, 27], [5, 10, 15, 29]])

m = np.array([False, False, True, True, False])

Consequently, I have:

data[m] -> array([[ 3, 6, 13, 25], [ 4, 8, 14, 27]])

and

data[m][:, 2:] -> array([[13, 25], [14, 27]])

I would like to assign a matrix to data[m][:, 2:]. Something like:

data[m][:, 2:] = np.array([[2,2], [2,2]])

and end up with the data like:

np.array([[1, 2, 11, 21], [2, 4, 12, 23], [3, 6, 2, 2], [4, 8, 2, 2], [5, 10, 15, 29]])

My use case is for a huge dataset where I cannot go cell by cell assigning values. Also, I know I can duplicate the mask to the number of columns and then make every value in those columns, but the ones I want to assign, into False and use that final mask over the data, but I am searching for a better solution.

Upvotes: 1

Views: 609

Answers (1)

hpaulj
hpaulj

Reputation: 231385

Because data[m] uses boolean indexing (selecting rows) the result is a copy, not a view. The subsequent assignment modifies that, not data. You need to combine the indexing into one.

I suggested using the indices of the True values in m:

In [205]: data[[2,3],2:]
Out[205]: 
array([[13, 25],
       [14, 27]])

In [208]: m.nonzero()
Out[208]: (array([2, 3]),)
In [209]: data[m.nonzero(),2:]
Out[209]: 
array([[[13, 25],
        [14, 27]]])

But m can be used directly, since it is just selecting rows:

In [210]: data[m,2:]
Out[210]: 
array([[13, 25],
       [14, 27]])

It's a little trickier to use boolean indexing with others (list or slices), so that's why I started with the list [2,3].

Upvotes: 1

Related Questions