Unique Values and Sets

np.unique returns each distinct value once, sorted. It can also count how often each value appears and where it first appears. NumPy's set functions combine two 1-D arrays the way Python sets do.

Unique values

The result is sorted and has no repeats.

import numpy as np

arr = np.array([11, 11, 12, 13, 14, 15, 16, 17, 12, 13, 11, 14, 18])
print(np.unique(arr))

Output:

[11 12 13 14 15 16 17 18]

Counts and first positions

return_counts=True adds how many times each value appears. return_index=True adds the position of its first appearance. With both, you get three arrays back, in the order values, positions, counts.

import numpy as np

arr = np.array([11, 11, 12, 13, 14, 15, 16, 17, 12, 13, 11, 14, 18])
values, first, counts = np.unique(arr, return_index=True, return_counts=True)
print(values)
print(first)
print(counts)

for m, p in zip(values, counts):
    if p > 1:
        print(m, "appears", p, "times")

Output:

[11 12 13 14 15 16 17 18]
[ 0  2  3  4  5  6  7 12]
[3 2 2 2 1 1 1 1]
11 appears 3 times
12 appears 2 times
13 appears 2 times
14 appears 2 times

Unique rows and columns

Without an axis, a 2-D array is flattened first. axis=0 keeps the unique rows and axis=1 the unique columns.

import numpy as np

m = np.array([[11, 11, 12, 11],
              [13, 11, 12, 11],
              [16, 11, 12, 11],
              [11, 11, 12, 11]])
print(np.unique(m))
print(np.unique(m, axis=0))
print(np.unique(m, axis=1))

Output:

[11 12 13 16]
[[11 11 12 11]
 [13 11 12 11]
 [16 11 12 11]]
[[11 11 12]
 [11 13 12]
 [11 16 12]
 [11 11 12]]

Set operations

union1d keeps values in either array. intersect1d keeps values in both. setdiff1d(a, b) keeps values of a that are not in b. setxor1d keeps values in exactly one. np.isin gives a mask of which values of a appear in b. Pass assume_unique=True only when neither array has repeats. It is faster, and wrong if repeats are present.

import numpy as np

a = np.array([1, 2, 3, 4])
b = np.array([3, 4, 5, 6])
print(np.union1d(a, b))
print(np.intersect1d(a, b, assume_unique=True))
print(np.setdiff1d(a, b))
print(np.setxor1d(a, b))
print(np.isin(a, b))

Output:

[1 2 3 4 5 6]
[3 4]
[1 2]
[1 2 5 6]
[False False  True  True]

Key takeaways

• np.unique returns sorted values with no repeats

• return_counts and return_index add counts and first positions

• axis=0 finds unique rows, axis=1 unique columns

• union1d, intersect1d, setdiff1d, and setxor1d work on 1-D arrays

• Next: Linear Algebra

• Previous: Sums, Products, and Rounding