mice-py: Multiple Imputation by Chained Equations in Python

A comprehensive Python implementation of Multiple Imputation by Chained Equations (MICE) for handling missing data in statistical analysis and machine learning workflows.

Key Features

Multiple Imputation Methods

Choose from five robust imputation strategies:

PMM (Predictive Mean Matching) - Maintains distributional properties
CART (Classification and Regression Trees) - Handles non-linear relationships
Random Forest - Captures complex interactions
MIDAS (Multiple Imputation with Distant Average Substitution) - Efficient for small samples
Sample - Simple random sampling from observed values

Flexible Configuration

Automatic predictor matrix estimation
Custom visit sequences for imputation order
Method-specific parameter control
Mixed data types (numeric and categorical)

Statistical Pooling

Rubin’s rules for combining estimates
Fraction of missing information (FMI)
Confidence intervals and standard errors
Formula-based model fitting with statsmodels integration

Diagnostic Tools

Convergence diagnostics (chain statistics)
Stripplots, box plots, and density plots
Missing data pattern visualization
XY plots for bivariate relationships

Quick Example

import pandas as pd
from imputation import MICE

# Load your data with missing values
df = pd.read_csv("your_data.csv")

# Initialize and run MICE
mice = MICE(df)
mice.impute(n_imputations=5, maxit=10, method='pmm')

# Fit a model and pool results
mice.fit('outcome ~ predictor1 + predictor2')
results = mice.pool(summ=True)
print(results)