Understanding numpy.argsort()

• An Array of Indices
• Why Use argsort Instead of Just Sorting?
• Sorting Higher-Dimensional Arrays
• Conclusion

The argsort method in numpy returns an array of indices that can be used to sort the array itself or another array. At first, its mechanism might feel a little confusing, but the idea is both powerful and straightforward once you understand how it works and why we need it.

An Array of Indices

Take this example:

import numpy as np

arr = np.array([10, 15, 3])
sorted_indices = np.argsort(arr)
print(sorted_indices)

This outputs:

[2 0 1]

• 2 means the smallest element (3) is at index 2.
• 0 means the next element in sorted order (10) is at index 0.
• 1 means the last element (15) is at index 1.

So instead of giving us the sorted array, argsort tells us the positions of the sorted values in the original array.

Why Use argsort Instead of Just Sorting?

You may wonder why not just call np.sort and be done with it? The difference is that argsort preserves order information. This becomes useful when you want to sort related arrays together.

For example:

fruits = np.array(['orange', 'apple', 'banana'])
prices = np.array([1.5, 2.5, 1.0])

We want to sort prices in ascending order and apply the same ordering to fruits:

indices = np.argsort(prices)
sorted_prices = prices[indices]
sorted_fruits = fruits[indices]
print(sorted_prices)
print(sorted_fruits)

This outputs:

[1.  1.5 2.5]
['banana' 'orange' 'apple']

Here, argsort gave us the index order to sort both arrays consistently. That way, prices and fruits stay aligned.

Sorting Higher-Dimensional Arrays

argsort works with arrays of any number of dimensions. By default, it sorts along the last axis (axis=-1), but you can specify another axis if needed.

Example with a 3D array:

arr_3d = np.array([
    [[3, 1, 2],
     [9, 6, 4]],
    [[7, 5, 8],
     [0, 2, 1]]
])

print(np.argsort(arr_3d, axis=-1))

It outputs:

[[[1 2 0]
  [2 1 0]]
 [[1 2 0]
  [0 2 1]]]

Each sub-array gets sorted independently along the last axis (columns, in this case).

Conclusion

numpy comes with many specialized methods that can feel unintuitive at first glance. But once you see their purpose, they become simple and powerful tools.

argsort is one of those methods: instead of just sorting values, it gives you the ordering information, which is invaluable when you need to sort related arrays or keep track of ranks and positions.