Winsorize Outlier Values

Winsorize Outlier Values

Winsorization replaces extreme values with the percentile boundary — rather than removing them. This preserves dataset size while limiting the influence of outliers on model training.

For lower percentile $p_{lo}$ and upper percentile $p_{hi}$:

1. Compute $v_{lo}=\text{percentile}(\text{arr},p_{lo})$ and $v_{hi}=\text{percentile}(\text{arr},p_{hi})$

2. Clip: any value below $v_{lo}$ becomes $v_{lo}$; any above $v_{hi}$ becomes $v_{hi}$

Example: arr = [1, 2, 3, 4, 100], lower_pct = 0, upper_pct = 75

Unlike z-score thresholding, winsorization makes no distributional assumption — it is purely rank-based.

Your task:

Implement winsorize(arr, lower_pct, upper_pct) using np.percentile and np.clip.