Day 2 — NumPy & Arrays | AI Engineering Internship

0–5 min
Real Hook

5–25 min
Core Concept

25–45 min
Live Coding

45–55 min
You Code It

55–60 min
Cheat Sheet

🔥 Real-World Hook (0–5 min)

Every image your phone camera takes is a 3D NumPy array: height × width × 3 (RGB). Instagram's filters? NumPy math on pixels. Face detection? NumPy arrays sliced and convolved.

Python loops over pixels would take 10 seconds per photo. NumPy vectorization does it in 0.001 seconds. That's why every AI framework is built on NumPy.

📈 Core Concepts (5–25 min)

💡 The 3 NumPy Ideas

Shape — dimensions of the array: (28, 28) = 28 rows, 28 cols.
Vectorization — operations apply to ALL elements at once (no loop needed).
Broadcasting — arrays with different shapes can combine: (3,) + (3,1) works automatically.

💻 Live Demo (25–45 min)

>import numpy as np

# 1. Shapes — think of arrays as tensors
pixel_row = np.array([255, 128, 0, 64, 200])   # 1D: shape (5,)
image     = np.zeros((28, 28))               # 2D: 28x28 grayscale
rgb_img   = np.zeros((28, 28, 3))            # 3D: 28x28 colour
print(image.shape, image.dtype)               # (28, 28) float64

# 2. Vectorization — no loop, math on all at once
scores = np.array([45, 78, 92, 60, 55])
normalized = (scores - scores.min()) / (scores.max() - scores.min())
print(normalized)                             # all values in [0.0, 1.0]

# 3. Broadcasting — align shapes automatically
# Subtract mean from each row (centering data)
data = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
row_means = data.mean(axis=1, keepdims=True)   # shape (3,1)
centered  = data - row_means                # broadcasts: (3,3) - (3,1)

# 4. Indexing and slicing
big = np.arange(100).reshape(10, 10)
patch = big[2:5, 3:7]                       # rows 2-4, cols 3-6
mask  = scores > 60                         # boolean mask
print(scores[mask])                           # [78 92]

🌎 Real-World

When you load an image with PIL/OpenCV, it becomes a NumPy array. Resize = array slice + interpolation. Brightness = array multiply. Grayscale = weighted sum across axis 2.

📋 You Code It (45–55 min)

✍ Task

Create a (5, 4) random array of student test scores (0–100). Compute: row-wise mean (per student), column-wise mean (per subject), then normalize the whole matrix to [0, 1].

⚡ Day 2 Cheat Sheet

np.array([...])

Create array from list

arr.shape / arr.dtype

Dimensions and type

np.zeros((m,n))

m×n zero matrix

np.random.rand(m,n)

Random floats [0,1)

arr.mean(axis=0)

Column-wise mean

arr.reshape(r, c)

Change shape (same data)

arr[mask]

Boolean indexing

arr1 @ arr2

Matrix multiplication

← Day 01 Day 03 · Pandas & EDA →