MRMCD2024 No More Loopy Code: Data Science Goes Functional

Functions in functional programming should be small, pure and declarative - making them easier to verify, test and debug

Data science workflows commonly involve chaining transformations like filtering, mapping and aggregating data - functional programming provides clean patterns for these operations

Using immutable state and avoiding global variables makes code more maintainable and less prone to errors

Map and filter are core functional programming concepts that allow operating on collections of data in a declarative way

Functions can be composed together into pipelines, making complex data transformations more readable and maintainable

Functional code is often easier to parallelize since pure functions can run independently without side effects

There’s no need to go “100% functional” - start by gradually introducing functional concepts that provide value

Strong type systems in functional programming help catch errors early in data transformations

Small, pure functions are easier to reason about in isolation compared to large imperative functions with side effects

While functional programming has a learning curve, it can make data science code more robust and maintainable in the long run