Daily variation analysis in Polars: Rolling std for negative values only The situation Hi there, I have a Polars DataFrame with columns like ['Date','category','number'], already grouped. This DataFrame contains almost all dates for 7 months (close to 25 million rows). I want to perform some analysis about the growth for each category. A simple way to start is by identifying days with significant positive or negative daily variations. What I want to calculate daily_variation %daily_variation rolling_mean_%daily_variation_30_days rolling_mean_%daily_std_30_days rolling_mean_%daily_std_30_days_neg_values ← This is where I'm stuck The problem I realize that Polars doesn't have a straightforward method for calculating a rolling std of only negative values (or I lack the knowledge). I've tried various approaches but can't figure out how to filter only negative values within a rolling window operation. How can I calculate a rolling std that only considers negative values within each 30-day window? Any help would be greatly appreciated! What I've tried # This works fine for regular rolling mean df = df.with_columns([ pl.col("daily_variation_pct").rolling_stdwindow_size=30).alias("rolling_std_30d") ]) # But I need something like this for negative values only: # df.with_columns([ # pl.col("daily_variation_pct").filter(< 0).rolling_std(30).alias("rolling_std_30d_neg") # ]) Continue reading...
