NumPy and Vectorization

Under development

This lesson is part of the course scaffold and is being actively written. The learning objectives and outline below define its final scope.

Learning objectives

By the end of this lesson you will be able to:

• Construct and manipulate NumPy arrays of prices and returns using appropriate dtypes, slicing, and views versus copies.
• Rewrite a loop-based indicator calculation as a vectorized, broadcast expression and benchmark the speedup.
• Diagnose floating-point precision failures (accumulation error, catastrophic cancellation, unsafe equality tests) in return calculations and apply numerically stable alternatives.
• Control NaN propagation in an array pipeline, choosing deliberately between nan-aware reductions and explicit masking.

Outline

1. Arrays, dtypes, and memory layout — why ndarray beats Python lists for market data
2. Indexing and slicing — views vs copies, boolean masks, fancy indexing
3. Broadcasting rules — aligning shapes without writing loops
4. Vectorization in practice — rewriting a loop-based signal, timing it honestly
5. Ufuncs and reductions — cumulative products, axis semantics, log-space computation
6. Floating-point pitfalls — precision limits, cancellation, comparing floats safely
7. NaN and missing data — propagation rules, nan-aware functions, masking strategies

Prerequisites

• None — this part starts the programming track.