CODEWITHSUNDEEP
pythonarrayslistnumpyindexing
S.A.
S.A.

Reputation: 2151

numpy indexing into array of dtype=object

I am creating an ndarray from ragged nested sequences, like this:

import numpy as np
arr = np.array([[1,2], [3,4,5], [6,7,8,9], [10]], dtype=object)

In a next step, I want to use a "boolean mask" of the same length as arr to get certain elements.

Passing a mask of all True works:

my_mask = [True, True, True, True]
arr[my_mask]
# array([[list([1, 2]), list([3, 4, 5]), list([6, 7, 8, 9]), list([10])]],
      dtype=object)

However, other masks don't seem to work:

my_mask = [True, False, True, True]
arr[my_mask]
# array([], shape=(0, 4), dtype=object)

Why does the above result in an empty array?

UPDATE: In the example above I wrote arr[my_mask], but the error I got locally was actually obtained via arr[True, False, True, True], which should rather be arr[[True, False, True, True]], note the double brackets. Thanks to @Ivan and @user1740577. As such, this is not unexpected behavior, but rather a user mistake during indexing.

Upvotes: 0

Views: 719

Answers (2)

Ivan
Ivan

Reputation: 40668

You should call arr[my_mask] instead of arr[True, False, True, True].

The reason is that by indexing with True you are adding a dimension,

>>> arr[True]
array([[list([1, 2]), list([3, 4, 5]), list([6, 7, 8, 9]), list([10])]],
      dtype=object)

but then mask the second axis (formerly the first axis) with False i.e. returning no elements.

Upvotes: 1

I'mahdi
I'mahdi

Reputation: 24049

instead of using:

arr[True, False, True, True]

using this (when you want to pass mask, pass array of mask):

arr[[True, False, True, True]]

Upvotes: 1

Related Questions