Polars

When to Use

• Performance: 10-100x faster than Pandas on large data.
• Lazy Evaluation: Define a query plan and execute it efficiently (

  .collect()

  ).
• Streaming: Process datasets larger than RAM.

Core Concepts

Lazy API (

.lazy()

)

Expressions

pl.col("a").sum()

Compatibility

Best Practices (2025)

• Use Lazy Mode:

  df.lazy().filter(...).collect()

  .
• Use

  scan_parquet

  : Don't load the whole file; scan it.
• Use GPU engine: Polars has experimental NVIDIA RAPIDS support.

• Don't treat it like Pandas: The API is different (

  axis=1

  doesn't exist; use expressions).

References

• Polars Documentation