pandas Series.rolling(50).mean() is ~2.74× slower than polars rolling_mean(50) for a 50-row rolling mean

Same window, same column, same row order.

Result

Run	pandas Series.rolling(50).mean()	polars rolling_mean(50)	Ratio
Run 1	12.98	4.59	2.83×
Run 2	12.36	4.56	2.71×
Run 3	16.32	5.96	2.74×

Spread over 3 runs: pandas Series.rolling(50).mean() CV 15.3% / polars rolling_mean(50) CV 15.8% / ratio CV 2.3%

df['v'].rolling(50).mean()

df['v'].rolling_mean(window_size=50)

400,000 float values in the order they were generated.

Rolling windows over grouped keys behave differently and are not measured here.

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