Advanced NumPy: Linear Algebra and Array Manipulation

Learning Objectives: After this lesson, you'll master NumPy's linear algebra capabilities, random number generation, array reshaping, and advanced indexing techniques essential for data science and machine learning.

Reshaping Arrays

Reshaping is fundamental to data manipulation—it lets you transform data between different dimensional representations without changing the underlying values.

Explore how arrays store elements and how operations like append and pop work:

When you reshape this 12-element array into a (3,4) matrix, the elements are distributed row by row:

Flatten and Ravel

Converting multi-dimensional arrays back to 1D:

Transpose and Axis Swapping

Stacking and Splitting Arrays

Stacking (Combining Arrays)

Splitting Arrays

Linear Algebra Operations

Linear algebra is the mathematical foundation of machine learning. NumPy provides comprehensive support. The diagram below shows common linear algebra workflows:

Matrix Multiplication

Dot Product of Vectors

Matrix Properties

Eigenvalues and Eigenvectors

Eigenvalues are crucial for understanding data transformations and dimensionality reduction (like PCA):

Solving Linear Systems

Solving Ax = b is fundamental to many algorithms:

Advanced Indexing

Using np.where()

np.where() is like a vectorized if-else statement:

np.select() for Multiple Conditions

np.clip() for Bounding Values

Random Number Generation

NumPy's random module is essential for simulations, sampling, and machine learning. Here's a visualization of different distribution types:

Random Number Generators

Sampling and Shuffling

Statistical Distributions

Practical Example: Data Preprocessing Pipeline

Let's combine everything into a realistic data preprocessing example:

Key Takeaways

✅ Reshaping transforms array dimensions with reshape(), flatten(), ravel(), and transpose

✅ Stacking and splitting combine or divide arrays with vstack(), hstack(), concatenate(), split()

✅ Linear algebra operations include matrix multiplication (@), inverse, determinant, eigenvalues, and solving systems

✅ Advanced indexing with np.where(), np.select(), np.clip() enables powerful conditional operations

✅ Random generation provides tools for sampling, shuffling, and various statistical distributions

✅ Data preprocessing pipelines combine these tools for real-world data preparation

Connections: Advanced NumPy in Practice

🔗 Connection to Machine Learning

🔗 Connection to Deep Learning

NumPy operations mirror neural network computations:

• Forward pass: Matrix multiplications (@ operator)
• Batch normalization: Mean/std along batch axis
• Dropout: Random masking
• Weight initialization: Random distributions

🔗 Connection to Statistics

Practice Exercises

Exercise 1: Matrix Operations

Exercise 2: Data Simulation

Next Steps

Now that you've mastered NumPy, you're ready for pandas—the library that builds on NumPy to provide intuitive data structures for real-world data analysis.

Ready to work with real datasets? Pandas is next!