NumPy — whole columns at once

You will be able to

• ✓ Contrast a Python list with a NumPy array
• ✓ Use vectorised arithmetic and aggregations
• ✓ Know dtype and broadcasting at a reading level

Lists are boxes; arrays are a grid of one type

A Python list can hold a number, a string and another list. That flexibility costs speed: every item is a separate object. A NumPy array is one type (float64, int32) laid out in a rectangle. temps + 1 adds one to every cell. temps[temps >= 35] is a filter without an explicit loop.

Shape matters. A column of 1440 minute-readings is shape (1440,). A day’s worth of 10 sensors is (1440, 10). Images are often (height, width, 3) for RGB. Once you see shape, every later tensor in PyTorch is the same idea.

Vectorisation is the habit

Write the maths as if it applied to one number, then let NumPy apply it to the array. fahrenheit = celsius * 9 / 5 + 32 works on an array. A Python loop that appends each conversion is what you write in this sandbox — and what you should stop writing on a laptop once the array is large.

Aggregations: np.mean, np.max, np.std, np.sum. Axis=0 vs axis=1 is “down columns” vs “across rows”. Lab reports live here: mean temperature per sensor, not a mystery number from Excel.

Broadcasting and dtype

Broadcasting lets a 1-D array of 10 offsets add to a (1440, 10) table — each column gets its calibration. When it fails, the error is about shape. Read the shapes; do not randomly .reshape until it “runs”.

dtype=float32 halves memory versus float64. For money, prefer decimals or integer paise — binary floats are the reason 0.1 + 0.2 looks ugly. For sensors, float64 is the default and usually fine.

Where NumPy sits in an application

Load with pandas or np.loadtxt, compute with NumPy, plot with matplotlib, store summaries in SQLite. Do not persist a 2 GB array as a CSV of 12 million lines if you can use .npy or a database of aggregates.

NumPy is not a table with column names. Named columns are pandas. Mixing them is normal: df["celsius"].to_numpy().

Words that matter

ndarray

NumPy’s N-dimensional array: one dtype, a shape, vectorised ops.

Vectorisation

Expressing an operation on the whole array instead of a Python loop.

dtype

The element type (int32, float64, …).

Broadcasting

Stretching smaller arrays to match a larger shape during arithmetic.

Common mistakes

On a full Python install — numpy

The standard numeric array library. pandas, scikit-learn, OpenCV and most science code expect it.

pip install numpy

Calibrate a column of °C, flag heat, report mean.

import numpy as np

celsius = np.array([29.0, 36.5, 33.0, 40.0, 31.2])
calibrated = celsius + 0.4
hot = calibrated[calibrated >= 35]
print("hot", hot)
print("mean", calibrated.mean())
print("peak", calibrated.max())

Example program — The same maths with a list (sandbox analog)

This is the loop NumPy replaces. Add 1 °C of calibration, keep values >= 35.

Your turn — Scale a column

readings = [10, 20, 30]. Build scaled where each value is times 2. Print so 40 appears (from 20×2).

Self-assessment

Check your understanding before you mark the lesson complete.