May 4, 2026

Why I reach for Polars over pandas

For most analytical workloads, the Polars API is closer to SQL than to pandas, and the speed-up is the smaller part of the win.

TODO — write the post body here. Outline:

  • The honest one-liner: Polars is faster, but the API is the reason I switched, not the benchmarks.
  • Show a side-by-side: same query in pandas (chained .groupby().agg().reset_index().merge()) vs Polars (one lazy expression that reads like SQL).
  • The lazy .collect() model — query optimizer pushes filters down, prunes columns, fuses ops. Why this matters at 100M rows.
  • Where pandas still wins: small data, plotting integration, certain time series operations, the long tail of libraries that expect a DataFrame.
  • Pragmatic recommendation: Polars for ETL and heavy aggregations, pandas for the last-mile analytical work where you're poking at a small slice.
All notes