Indexing and slicing · NumPy

One dimension

Python

import numpy as np

a = np.array([10, 20, 30, 40, 50, 60])
print(a[0], a[-1])       # first and last: 10 60
print(a[1:4])            # positions 1, 2, 3: [20 30 40]
print(a[:3])             # the first three
print(a[3:])             # from position 3 to the end
print(a[::2])            # every second element: [10 30 50]
print(a[::-1])           # reversed

Slices work like Python's: start:stop:step, where stop is not included.

Two dimensions

Python

import numpy as np

m = np.array([[ 1,  2,  3,  4],
              [ 5,  6,  7,  8],
              [ 9, 10, 11, 12]])

print(m[1, 2])        # row 1, column 2: 7
print(m[0])           # the whole first row
print(m[:, 1])        # the whole second column: [ 2  6 10]
print(m[0:2, 1:3])    # rows 0-1, columns 1-2: a 2 x 2 block
print(m[-1, -1])      # bottom-right: 12

Index with [row, column]; slices use start:stop for each axis, and : alone means everything along that axis.

Use one pair of brackets with a comma, m[row, column]. Inside, : on its own means "everything along this axis".

Choosing several positions

Python

import numpy as np

scores = np.array([72, 95, 88, 61, 79])
print(scores[[0, 2, 4]])        # the elements at positions 0, 2 and 4

m = np.arange(1, 13).reshape(3, 4)
print(m[[0, 2]])                # rows 0 and 2
print(m[:, [3, 0]])             # columns 3 and 0, in that order

A list of indexes, often called fancy indexing, picks exactly those positions, in the order you give.

Where's the biggest?

Python

import numpy as np

temps = np.array([21.5, 24.0, 19.8, 26.3, 23.1])
days = np.array(["Mon", "Tue", "Wed", "Thu", "Fri"])

hottest = temps.argmax()            # the POSITION of the maximum
print(hottest, days[hottest], temps[hottest])
print(days[temps.argmin()])         # the coolest day
print(days[np.argsort(temps)])      # days from coolest to hottest

argmax and argmin return positions, not values, so you can use them to index a second, related array.

Slices are views

Python

import numpy as np

a = np.array([1, 2, 3, 4, 5])
part = a[1:4]          # a view: it shares memory with a
part[0] = 99
print(a)               # [ 1 99  3  4  5]: the original changed

b = np.array([1, 2, 3, 4, 5])
copy = b[1:4].copy()   # an independent copy
copy[0] = 99
print(b)               # unchanged

A slice doesn't copy the data: it's a view onto the same memory. That makes slicing fast even on huge arrays, but changing the view changes the original. Call .copy() when you need an independent array. (Fancy indexing with a list always makes a copy.)

Changing values through an index

Python

import numpy as np

m = np.zeros((3, 4), dtype=int)
m[0] = 1             # a whole row
m[:, -1] = 9         # the last column
m[1:, 1:3] = 5       # a block
print(m)