robustats-py 0.2.0

Robust statistics for NumPy: replaces mean with median for outlier-resistant analysis

robustats

Robust statistics for NumPy — replaces mean with median for outlier-resistant data analysis.

Why?

NumPy's np.mean is sensitive to outliers. One extreme value can distort your entire analysis. robustats swaps mean with median by default — a far more robust estimator — while keeping the same familiar API.

data = [1, 2, 3, 4, 1000]

np.mean(data)      # 202.0  ← distorted by outlier
rs.mean(data)      #   3.0  ← robust, ignores outlier

Installation

pip install robustats

Quick Start

import numpy as np
import robustats as rs

data = [1, 2, 3, 4, 1000]

# Drop-in replacements
rs.mean(data)               # 3.0    (median, robust)
rs.nanmean(data)            # 3.0    (nanmedian, robust)
rs.std(data)                # 1.4826 (MAD-based, robust)
rs.var(data)                # 2.198  (MAD², robust)
rs.mad(data)                # 1.4826 (Median Absolute Deviation)

# Use arithmetic mean explicitly
rs.mean(data, use_median=False)    # 202.0

# Weighted median
rs.average(data, weights=[1,1,1,1,10])   # 5.0

Context Manager — patch np.mean temporarily

with rs.robust_mode():
    np.mean(data)     # 3.0 inside the block
np.mean(data)         # 202.0 restored after the block

Global Patch — replace np.mean everywhere

rs.patch_numpy()
np.mean(data)     # 3.0  (global replacement)

rs.unpatch_numpy()
np.mean(data)     # 202.0 (restored)

API Reference

Function	Description
rs.mean(a, axis, use_median=True)	Median (or mean) of array
rs.nanmean(a, axis, use_median=True)	Nanmedian (or nanmean)
rs.std(a, axis, use_mad=True)	MAD-based std (or numpy std)
rs.var(a, axis, use_mad=True)	MAD²-based var (or numpy var)
rs.mad(a, axis, scale=1.4826)	Median Absolute Deviation
rs.average(a, weights, use_median=True)	Weighted median (or average)
rs.robust_mode()	Context manager patching np.mean
rs.patch_numpy()	Globally replace np.mean / np.nanmean
rs.unpatch_numpy()	Restore originals

When to use robustats

• Your data may have outliers or measurement errors
• You're doing exploratory data analysis on real-world datasets
• You need a quick drop-in without changing all your np.mean calls (robust_mode context)
• You want robust standard deviation via MAD instead of classical std

License

MIT