Pyra

Pyra is a static analysis tool designed specifically for data science applications written in Python. It extends the Lyra framework to identify subtle bugs and anti-patterns that can lead to misleading results, reproducibility issues, and data leakage in machine learning pipelines.

Key Features

• Abstract Datatype Analysis — Performs static analysis to track data types and operations throughout Python code
• Data Leakage Detection — Identifies when data preprocessing (like scaling) occurs before train-test splits
• Reproducibility Checks — Flags missing random seeds in operations like train_test_split
• API Misuse Detection — Catches inappropriate method usage like inplace=True assignments or incorrect plot types
• Visual Output — Generates PDF control flow graphs annotated with analysis results

Philosophy

Pyra aims to bring formal verification techniques to data science workflows, helping developers write more robust and reproducible code through static analysis rather than runtime testing.