pandas boolean mask is ~8.13× slower than polars filter for filtering rows by a float threshold

Keeps roughly half of the rows.

Result

Spread over 3 runs: pandas boolean mask CV 9.4% / polars filter CV 2.6% / ratio CV 8.7%

The two calls

df[df['v'] > 0.5]

df.filter(pl.col('v') > 0.5)

Conditions

400,000 rows, float column, threshold 0.5.

Where this number does not apply

A highly selective filter (1% of rows) has a different cost because the copy dominates.

Reproduce

cd bench && python run.py --run 1 --out ../data/run1.json