Stephen Macke - Python as a Hackable Language for Interactive Data Science | PyData Global 2023

Python’s dominance in data science stems from its extensive ecosystem of data-related libraries and excellent tooling for interactive programming

The IPyFlow kernel enables reactive execution in Jupyter notebooks, automatically refreshing dependent cells when variables change without unnecessary recalculations

AST (Abstract Syntax Tree) transformations allow Python to be extended with new features like top-level await and optional chaining, making it more convenient for specific use cases

Memoization can be used to skip expensive computations when inputs haven’t changed, making notebook interactions appear instantaneous

Piccolo library enables composable AST transformations that don’t conflict with each other, unlike traditional AST transformers

Interactive widgets (like sliders) can be integrated with reactive execution for dynamic data visualization and educational purposes

Out-of-order execution and state management issues in notebooks can be addressed through proper instrumentation and dependency tracking

Static code analysis alone isn’t sufficient for determining dependencies in notebooks - runtime introspection is often necessary

The combination of reactive execution and memoization can significantly improve the interactive data analysis workflow by reducing wait times

Python’s extensibility enables creating tools that enhance the notebook experience while maintaining a clean, intuitive syntax