Polars

If you're hitting pandas performance walls but your data still fits in RAM, this is the obvious next step. Built on Apache Arrow with lazy evaluation and parallel execution by default, it's genuinely 5-10x faster for typical operations on datasets between 1GB and 100GB. The expression API is cleaner than pandas once you adjust to no index and strict typing. Use `scan_csv` instead of `read_csv` for large files to get automatic query optimization and predicate pushdown. The skill covers the core concepts well, including when to go lazy versus eager and how window functions actually work with `over()`. Just remember it won't help if your data exceeds RAM, that's when you need dask or vaex instead.

Install to Claude Code

npx -y skills add k-dense-ai/claude-scientific-skills --skill polars --agent claude-code