0

I've got a manual way to reference the numpy array, but can't figure out how to do it in a numpy-like manner.

I am looking for a numpy-like equivalent of the following line: 

np.array([arr[0,index[0]], arr[1,index[1]], arr[2,index[2]]]).reshape(3,1)

Here is the code (simplified to pinpoint the issue):

arr = np.arange(9).reshape(3,3)
index = np.array([2,0,1])
print(arr)
result = np.array([arr[0,index[0]], arr[1,index[1]], arr[2,index[2]]]).reshape(3,1)
print(result)

it produces the following: 

[[0 1 2]
 [3 4 5]
 [6 7 8]]
[[2]
 [3]
 [7]]

I am looking for numpy-style code to achieve what the line 

np.array([arr[0,index[0]], arr[1,index[1]], arr[2,index[2]]]).reshape(3,1)

is doing here, for example: 

arr[:,index[:]]

Obviously, this doesn't work correctly, because index[:] denotes the whole row instead of picking just one corresponding value.

This must be very simple, I am just somehow stuck on it. Can't figure out how to do it without a long line that manually brings every index for every row. In my case I have thousands of rows and thousands of columns, so definitely can't use the manual way.

Improve this question
1
  • @Divakar I've gone through the referenced posts, and came up with the answer to my question: arr[[0,1,2],index] does exactly what I need. But I still can't understand how it works, why it does what it does. I am new to numpy and would greatly appreciate some help. More specifically, how exactly numpy is treating 2 totally different instructions which look very similar: 1) arr[0:3,index] and 2) arr[[0,1,2],index] Commented Aug 25, 2019 at 18:09

2 Answers 2

Reset to default
1

Probably the simplest way to achieve what you are going for is by direct indexing:

arr = np.arange(9).reshape(3,3)
index = np.array([2,0,1])
result = arr[range(index.shape[0]), index].reshape(index.shape[0], 1)

To answer your comment, the reason why slices don't produce the same results, i.e. code like this

arr[0:index.shape[0], index]

is due to the fact that the 0:index.shape[0] is a slice, not a list of indices to pick. Whenever you use slices as index, it understands as all rows in range or all columns in range. However if you list the indices instead (via lists or arrays) it will only pick the elements from the list.

There are many more tricks with NumPy indexing, their documentation is always a good source of information: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html.

Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

1

With a row selection of [0,1,2] rows and [2,0,1] columns there two ways of selecting items:

block
items [2,0,1] from each of the rows [0,1,2]

paired
items [0,2], [1,0], [2,1]

In MATLAB if you index arr([0,1,2],[2,0,1]) you get the block. In numpy indexing arr[[0,1,2],[2,0,1]] does the paired indexing that you want.

To get the block in numpy you can do arr[:, [2,0,1]] or arr[[[0],[1],[2]], [2,0,1]].

This second case actually makes a (3,1) array that broadcasts with a (3,) to create a (3,3) set of index pairs. To understand advanced indexing with more than one list or array, you need to understand broadcasting.

In the case of 1d arrays/lists that match in size the broadcasting is trivial - just pair them up:

arr([0,1,2],[2,0,1])
arr[0,2], arr[1,0], arr[2,1]
Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.