missmixed 1.1.0

An Adaptive, Extensible and Configurable Multi-Layer Framework for Iterative Missing Value Imputation

MissMixed

A Configurable Framework for Iterative Missing Data Imputation

MissMixed is a Python library designed for flexible and modular imputation of missing values in tabular datasets. It supports a wide range of imputation strategies, including ensemble methods, trial-based model selection, and deep learning integration — all within a customizable iterative architecture.

🔍 What is MissMixed?

MissMixed is not just a single algorithm — it’s a framework for building iteration-wise, model-aware imputation pipelines. It enables users to:

• Handle continuous, categorical, or mixed-type features
• Define custom model configurations at each iteration
• Combine multiple imputation algorithms (e.g., RandomForest, KNN, Deep Neural Networks)
• Dynamically evaluate and update imputed values using internal validation

Whether you’re working with low-dimensional medical data or large-scale mixed-type datasets, MissMixed is designed to offer accuracy, adaptability, and interpretability.

🚀 Installation

pip install missmixed

📦 Requirements

• Python ≥ 3.10
• NumPy
• Pandas
• scikit-learn
• XGBoost
• TensorFlow or Keras (for deep model imputation)
• tqdm

Dependencies will be installed automatically via pip.

📖 Usage

See the example folder for how to define: Custom Iteration Architectures Mixed-type pipelines Trial-based imputation workflows

OR

Use Command-Line Interface (CLI)

 missmixed --path .\input_data.csv

💻 MissMixed CLI Options

The following table lists all command-line arguments for MissMixed:

Argument	Short	Type	Default	Description
--path	-p	str	required	Path to the input data file (CSV or XLSX).
--column	-col	str (list)	None	Specify categorical or non-categorical columns by name. Format: <type> <col1> <col2> ... • <type> must be cat / categorical or non-cat / non-categorical. • If no columns are listed after <type>, all columns are treated according to <type>. Example: --column cat age city → treat age and city as categorical. --column non-cat income → treat all except income as categorical.
--index	-idx	int (list)	None	Specify categorical or non-categorical columns by index (0-based). Format: <type> <idx1> <idx2> ... • <type> must be cat / categorical or non-cat / non-categorical. • If no indices are listed after <type>, all columns are treated according to <type>. Example: --index cat 0 2 4 → treat columns 0, 2, 4 as categorical. --index non-cat 1 3 → treat all except columns 1, 3 as categorical.
--initial-strategy	-s	str	mean	Initial strategy for filling NaN values. Choices: mean, median, most_frequent.
--metric	-m	str	r2_accuracy	Metric for model evaluation. Choices: r2_accuracy, mse.
--trials	-t	int	1	Number of trials for training imputers through all iterations.
--train-size	-ts	float	0.9	Train size ratio (validation size = 1 - train_size).
--verbose	-v	int	0	Verbosity level: 0 (silent), 1 (default), 2 (detailed).
--output	-o	str	imputed_data.csv	Path to save the imputed output file. Supports .csv or .xlsx.

📄 License

MIT License

📣 Citation

[1] M. M. Kalhori, M. Izadi, “A Novel Mixed-Method Approach to Missing Value Imputation: An Introduction to MissMixed”, 29th International Computer Conference, Computer Society of Iran (CSICC) – IEEE, 2025.

[2] M. M. Kalhori, M. Izadi, F. Akbari “MissMixed: An Adaptive, Extensible and Configurable Multi-Layer Framework for Iterative Missing Value Imputation”, IEEE Access, 2025 (under review).