MissMecha

MissMecha is a Python package for the systematic simulation, visualization, and evaluation of missing data mechanisms.
It provides a unified, principled interface to generate, inspect, and analyze missingness — supporting research, benchmarking, and education.

Documentation: https://echoid.github.io/MissMecha/

Highlights

• All About Missing Mechanisms

  • Simulate MCAR, MAR, and MNAR with flexible configuration
  • Currently supports:
    • 3× MCAR strategies
    • 8× MAR strategies
    • 6× MNAR strategies
    • Experimental support for categorical and time series missingness

• Missingness Pattern Visualization

  • Visual tools to inspect missing patterns and detect possible mechanism types (MCAR, MAR, MNAR)

• Flexible Generator Interface

  • Column-wise or global simulation
  • Scikit-learn style API (fit, transform)
  • Customize missing rates, dependencies, or simulate label-dependent missingness

• Evaluation Toolkit

  • Evaluate imputation with RMSE, MAE, accuracy, or hybrid AvgERR metric
  • Built-in statistical test: Little’s MCAR test

• SimpleSmartImputer

  • Lightweight, automatic imputer that detects column types
  • Mean for numerical columns, Mode for categorical columns, with verbose reporting

• Custom Mechanism Support (New in v0.1.2)

  • Now supports user-defined missing mechanisms via custom_class
  • Easily plug in your own masker with fit + transform interface
  • See Custom Mechanism Demo

• Improved MNAR Type 1 (New in v0.1.2)

  • Supports missing_rate-based quantile masking for fine control
  • E.g. mask top 30% values in continuous columns via missing_rate=0.3
  • Fixes limitations when types vary across columns
  • Thanks @mahshidkhatiri for raising the issue

Motivation

Working with missing data often means dealing with fragmented, inconsistent tools.

MissMecha solves this by offering a unified, reproducible, and flexible framework for simulating and analyzing missingness — covering the full range of MCAR, MAR, and MNAR patterns.

Whether you're exploring datasets, designing controlled experiments, or teaching statistics —
MissMecha brings structure and clarity to missing data problems.

Quick Preview

from missmecha import MissMechaGenerator
import numpy as np

X = np.random.rand(100, 5)

generator = MissMechaGenerator(
    mechanism="mar", mechanism_type=1, missing_rate=0.3
)
X_missing = generator.fit_transform(X)

Or configure different mechanisms for each column:

generator = MissMechaGenerator(
    info={
        0: {"mechanism": "mcar", "type": 1, "rate": 0.3},
        1: {"mechanism": "mnar", "type": 2, "rate": 0.4}
    }
)
X_missing = generator.fit_transform(X)

Watch a 5-minute live demo here: MissMecha: Flexible Missing Data Simulation (Vimeo)

Documentation & Demos

• Explore full API and tutorials: https://echoid.github.io/MissMecha/

Installation

pip install missmecha-py

Available on PyPI under the package name missmecha-py.

Author

Developed by Youran Zhou, PhD Candidate @ Deakin University

License

MIT License