Boolean masks · NumPy

A comparison on an array gives an array of True and False, one per element. That boolean array, a mask, is how NumPy filters data without loops.

Masks

Python

import numpy as np

scores = np.array([72, 95, 88, 61, 79, 54])
passed = scores >= 70
print(passed)                  # [ True  True  True False  True False]
print(scores[passed])          # only where the mask is True
print(scores[scores < 60])     # usually written in one go

Counting and testing

Python

import numpy as np

scores = np.array([72, 95, 88, 61, 79, 54])
print((scores >= 70).sum())        # True counts as 1: 4 passed
print((scores >= 70).mean())       # the fraction that passed
print((scores > 90).any())         # is at least one above 90?
print((scores > 50).all())         # is every score above 50?

True behaves like 1 and False like 0, so sum counts matches and mean gives the fraction.

Combining conditions

Python

import numpy as np

ages = np.array([15, 22, 37, 45, 63, 71, 19])
print(ages[(ages >= 18) & (ages < 65)])     # and
print(ages[(ages < 18) | (ages >= 65)])     # or
print(ages[~(ages >= 18)])                  # not

Use &, | and ~, not and, or and not, and put brackets around every condition: & binds more tightly than >=, so ages >= 18 & ages < 65 would be evaluated wrongly.

np.where: choose between two values

Python

import numpy as np

scores = np.array([72, 95, 88, 61, 79, 54])
print(np.where(scores >= 70, "pass", "fail"))

temps = np.array([-3, 5, 12, -8, 0])
print(np.where(temps < 0, 0, temps))      # negative readings become 0

print(np.where(scores >= 90))             # with one argument: the positions where it's True

np.where(condition, if_true, if_false) builds a new array element by element.

Replacing values

Python

import numpy as np

readings = np.array([12.5, -999.0, 13.1, 12.8, -999.0, 14.0])   # -999 means "sensor failed"
readings[readings == -999.0] = np.nan                           # mark them missing
print(readings)
print(np.nanmean(readings))                                     # the mean, ignoring missing values

marks = np.array([105, 88, -4, 72, 99])
print(np.clip(marks, 0, 100))                                   # force values into 0..100

Assigning to a masked selection changes those elements in place. np.nan means "not a number", the standard marker for missing values, and the nan... functions such as np.nanmean skip over it.

Masks on 2D arrays

Python

import numpy as np

sales = np.array([[120,  80, 150],
                  [ 90, 200,  60],
                  [ 30, 110, 170]])
print(sales[sales > 100])                 # matching values, flattened into 1D
print(np.argwhere(sales > 150))           # their (row, column) positions
big_rows = sales.sum(axis=1) > 320
print(sales[big_rows])                    # whole rows whose total is over 320