Indexing and Slicing

NumPy arrays power everything from simple calculations to machine learning models. Mastering indexing and slicing is your key to unlocking their full potential — it's how you grab exactly the data you need, fast and efficiently. 🔍

We'll start simple and gradually build to advanced, real-world techniques. By the end, you'll handle complex multi-dimensional data like a pro! 🐍✨

Creating NumPy Arrays - Your Foundation

Always start with:

import numpy as np

1D, 2D, and 3D Arrays

# 1D: Like a list or row
arr_1d = np.array([10, 20, 30, 40, 50])

# 2D: Like a table
arr_2d = np.array([
    [1, 2, 3],
    [4, 5, 6],
    [7, 8, 9]
])

# 3D: Like a stack of tables (e.g., RGB images)
arr_3d = np.array([
    [[101, 102, 103], [104, 105, 106]],  # Layer 1
    [[201, 202, 203], [204, 205, 206]],  # Layer 2
    [[301, 302, 303], [304, 305, 306]]   # Layer 3
])

print("3D shape:", arr_3d.shape)  # (3, 2, 3)

3D shape: (3, 2, 3)

🟢 DO: Use higher dimensions early — real data (images, videos, sensor readings) is rarely just 2D!

Basic Indexing - Single Elements

Zero-based indexing, just like Python lists.

1D and 2D

print(arr_1d[0])      # 10
print(arr_1d[-1])     # 50 (last)

print(arr_2d[1, 2])   # Row 1, Col 2 → 6

3D Indexing (layer, row, column)

print(arr_3d[2, 0, 1])  # Layer 2, Row 0, Col 1 → 302

🟡 Tip: Always check array.shape first — it tells you the exact order of dimensions!

Basic Slicing - Sub-Arrays

Format: start:end:step (end exclusive)

# 1D examples
print(arr_1d[1:4])     # [20 30 40]
print(arr_1d[::2])     # [10 30 50] (every second)
print(arr_1d[::-1])    # [50 40 30 20 10] (reverse)

# 2D examples
print(arr_2d[0:2, 1:])  # Rows 0-1, columns 1 to end

[[2 3]
 [5 6]]

Advanced Slicing with Steps and Negative Indices

# Reverse rows but keep columns normal
print(arr_2d[::-1, :])

# Every second element in both dimensions
print(arr_2d[::2, ::2])

[[7 8 9]
 [4 5 6]
 [1 2 3]]

[[1 3]
 [7 9]]

Real-World Example: Student Performance Dashboard

Let's build a complete analysis using only NumPy — no Pandas yet!

students = np.array(['Alice', 'Bob', 'Charlie', 'David', 'Eva'])

# Shape: (5 students, 3 subjects, 2 metrics: marks + attendance)
data = np.array([
    [[85, 95], [88, 90], [92, 94]],  # Alice: Math, Science, English
    [[90, 87], [76, 85], [85, 88]],
    [[78, 92], [92, 90], [88, 85]],
    [[92, 88], [85, 86], [79, 90]],
    [[88, 96], [90, 92], [94, 95]]
])

marks = data[:, :, 0]      # All students, all subjects, marks only
attendance = data[:, :, 1] # Attendance

print("Marks:\n", marks)
print("Attendance:\n", attendance)

Marks:
 [[85 88 92]
 [90 76 85]
 [78 92 88]
 [92 85 79]
 [88 90 94]]
Attendance:
 [[95 90 94]
 [87 85 88]
 [92 90 85]
 [88 86 90]
 [96 92 95]]

Complex Calculations

# Total and average marks per student
total_marks = marks.sum(axis=1)
avg_marks = marks.mean(axis=1)

# Average attendance per subject
avg_attendance_per_subject = attendance.mean(axis=0)

print("Total marks:", total_marks)
print("Average marks:", avg_marks)
print("Avg attendance per subject:", avg_attendance_per_subject)

Total marks: [265 251 258 256 272]
Average marks: [88.33333333 83.66666667 86.         85.33333333 90.66666667]
Avg attendance per subject: [91.4 88.6 90.4]

Conditional Grading (np.where)

grades = np.where(avg_marks >= 90, 'A',
              np.where(avg_marks >= 80, 'B', 'C'))

print("Grades:", grades)

Grades: ['B' 'C' 'B' 'B' 'A']

Ranking Students

ranks = (-total_marks).argsort().argsort() + 1  # Descending rank
print("Ranks:", ranks)

Ranks: [2 5 3 4 1]

🟢 DO: Use argsort() twice for stable ranking — it's a pro trick!

Boolean Indexing - Powerful Filtering

# Students with average >= 88
top_students = students[avg_marks >= 88]
print("Top performers:", top_students)

# Marks of students with low attendance in any subject (< 90)
low_att = students[np.any(attendance < 90, axis=1)]
print("Need attendance warning:", low_att)

Top performers: ['Alice' 'Eva']
Need attendance warning: ['Bob' 'Charlie' 'David']

Fancy Indexing - Irregular Selections

# Marks of Alice, Charlie, Eva (indices 0,2,4)
selected = marks[[0, 2, 4]]
print("Selected marks:\n", selected)

# Swap two students' data
marks[[0, 1]] = marks[[1, 0]]  # Swap Alice and Bob
print("After swap:\n", marks[0])  # Now shows Bob's original

Views vs Copies - Critical for Large Data

sub = marks[:2]      # View!
sub[0, 0] = 999
print("Original changed:", marks[0, 0])  # 999

safe = marks[:2].copy()  # Independent copy
safe[0, 0] = 111
print("Original safe:", marks[0, 0])     # Still 999

🔴 DON'T: Assume slices are copies — they are views! Always .copy() when you want independence.

Advanced Real-World: Image Cropping Simulation

# Simulate a 100x100 RGB image
img = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)

# Crop center 50x50 region
center_crop = img[25:75, 25:75]

# Flip horizontally
flipped = img[:, ::-1]

# Extract only red channel
red_channel = img[:, :, 0]

print("Center crop shape:", center_crop.shape)

Center crop shape: (50, 50, 3)

Beginner Mistakes to Avoid

🔴 DON'T: Use single index on multi-dim arrays when you need a slice — it reduces dimensions unexpectedly.

🔴 DON'T: Forget copy() in data pipelines — bugs from shared memory are hard to trace.