Day 02 — Part 1: NumPy Fundamentals

Agenda & Learning Roadmap

What You Are Walking Into

NumPy is not just a library. It is the memory model, the speed contract, and the coordinate system that every serious numerical tool in Python is built on. Pandas DataFrames sit on NumPy arrays. Scikit-learn models expect NumPy arrays. TensorFlow and PyTorch mimic NumPy's API. The hour you spend truly understanding NumPy pays compound interest for your entire data science career.

---

Today's Roadmap

NumPy Fundamentals
│
├── 01 → Introduction to NumPy
│         Why Python lists fail at scale | The ndarray concept
│         Memory layout | dtype system and why it matters
│
├── 02 → Array Creation
│         np.array | zeros/ones/full | arange vs linspace (when each)
│         Random module | eye/diag | zeros_like | reshape
│
├── 03 → Indexing & Slicing
│         1D / 2D / 3D indexing | Negative indices | Step slicing
│         Boolean indexing (the workhorse) | Fancy indexing | np.where
│         Views vs Copies — the silent bug factory
│
├── 04 → Vectorization
│         For-loop vs vectorized timing | Universal functions (ufuncs)
│         Axis-wise aggregation | Why it is fast (SIMD, cache locality)
│
├── 05 → Broadcasting
│         The 3 rules | Visual intuition | Row and column broadcasting
│         Practical patterns | Shape errors and how to read them
│
├── 06 → Exercises
│         Warm-up → Main → Stretch — realistic data science problems
│
└── 07 → Cheat Sheet
          Dense quick-reference with code for every operation

---

Time Allocation

#	Topic	Duration	Type
01	Introduction to NumPy	20 min	Lecture
02	Array Creation	25 min	Lecture + Code
03	Indexing & Slicing	30 min	Lecture + Code
04	Vectorization	20 min	Lecture + Demo
05	Broadcasting	25 min	Lecture + Visuals
06	Exercises	40 min	Hands-on Practice
07	Cheat Sheet Review	10 min	Review

Total: ~2.5 hours

---

Learning Objectives

By the end of this session, you will be able to:

• Explain why NumPy arrays are faster than Python lists and what makes that possible at the hardware level
• Create arrays using the right function for the situation — not just the one you remember first
• Access and modify array data using indexing, slicing, boolean masks, and fancy indexing
• Predict whether an operation will return a view or a copy, and explain why this matters
• Replace explicit for-loops with vectorized operations and measure the speedup
• Apply broadcasting to operate on differently-shaped arrays without writing extra code
• Read a broadcasting error message and fix the shape mismatch

---

Prerequisites

• Python basics: variables, lists, loops, functions
• Python installed with NumPy (pip install numpy or conda install numpy)
• A way to run code: Jupyter Notebook, VS Code, or a Python script

---

Setup Check

Run this before the session:

import numpy as np
print(np.__version__)  # 1.24 or higher is fine

Version Note

NumPy 1.24+ and 2.x are both fine for everything in these notes. The np.random.default_rng() API used here requires 1.17+.

---

Files In This Module

File	Topic
01-introduction-to-numpy.md	Why NumPy, the ndarray, memory layout, dtype
02-array-creation.md	Every way to build arrays, when to use each
03-indexing-and-slicing.md	Accessing data, views vs copies, boolean indexing
04-vectorization.md	Fast math without loops, ufuncs, axis operations
05-broadcasting.md	Operating on different-shaped arrays with visual intuition
06-numpy-exercises.md	Practice problems — warm-up, main, stretch
07-cheat-sheet.md	Dense quick-reference

---

How to Use These Notes

Study Strategy

Type every code example yourself — do not paste. The mistakes you make while typing teach you more than reading does. After each section, close the notes and try to reproduce the key ideas from memory. Do the exercises before checking solutions. Struggling for ten minutes before looking at the answer is worth more than reading it cold.

---

01-introduction-to-numpy