Python — NumPy

Install & import

pip install numpy

import numpy as np          # the universal alias

Why NumPy exists

Python lists do math with loops:

prices = [10.0, 20.0, 30.0]
doubled = [p * 2 for p in prices]     # slow loop, verbose

NumPy arrays do whole-array math at C speed:

a = np.array([10.0, 20.0, 30.0])

a * 2            # array([20., 40., 60.])  — one expression!
a + 5            # adds to every element
np.sqrt(a)       # element-wise square roots

This "apply the operation everywhere" behavior is called vectorization — NumPy's entire reason for being.

Creating arrays

np.array([1, 2, 3])                    # from a list
np.zeros((3, 4))                       # 3×4 of 0.0
np.ones(5)
np.arange(0, 10, 2)                    # like range: 0,2,4,6,8
np.linspace(0, 1, 5)                   # 5 evenly spaced: [0, .25, .5, .75, 1]
np.random.randint(1, 7, size=(3, 3))   # random matrix (dice rolls)

Shape & dimensions

m = np.array([[1, 2, 3],
              [4, 5, 6]])

m.shape        # (2, 3) — rows, columns
m.ndim         # 2
m.size         # 6
m.dtype        # int64 — one type per array

Indexing & slicing

m[1, 2]        # row 1, col 2 → 6
m[0]           # first row
m[:, 0]        # ALL rows, column 0 → [1, 4]
m[m > 2]       # boolean mask → [3, 4, 5, 6]  ← filtering by condition!

That last line — filtering an array by a comparison — is NumPy's signature move.

Aggregations

a = np.array([21.5, 22.1, 19.8])

a.mean(); a.sum(); a.min(); a.max()
a.std()                  # standard deviation
m.sum(axis=0)            # per-column sums
m.sum(axis=1)            # per-row sums

Real example — grade analysis

scores = np.array([88, 92, 79, 95, 67])

scores.mean()                     # class average
scores - scores.mean()            # each student vs average
np.where(scores >= 90, "A", "B")  # conditional per element
len(scores[scores < 80])          # count below 80 → 2

Broadcasting preview

Arrays of different shapes still combine when compatible:

m = np.array([[1, 2], [3, 4]])
row = np.array([10, 20])

m + row        # [[11,22],[13,24]] — row added to EVERY row

Gotchas: one wrong-typed value upcasts everything to float · a[1:] slices are VIEWS (mutating them changes a!) · == on arrays returns an array of booleans, not one bool.

Mini Practice

1. Vectorize a list-comprehension you wrote earlier; time both.
2. Dice-roll matrix 1000×6; compute per-column means.
3. Boolean-mask filter temperatures above average.
4. np.where grades: ≥90 A, ≥80 B, else C.
5. Prove the view-mutation gotcha; fix with .copy().

Next: pandas →