Clustering-Informed Shared-Structure VAE for Imputation
Create or reuse a CISSVAE Python virtual environment
Create Missingness Proportion Matrix
Autotune CISS-VAE hyperparameters with Optuna
Check PyTorch device availability
Cluster-wise Heatmap of Missing Data Patterns
Cluster Samples Based on Missingness Proportions
Cluster on Missingness Patterns
Cluster-wise summary table using a separate cluster vector (gtsummary ...
Compute per-cluster and per-group performance metrics (MSE, BCE)
Plot VAE Architecture Diagram
Run the CISS-VAE pipeline for missing data imputation
Implements the Clustering-Informed Shared-Structure Variational Autoencoder ('CISS-VAE'), a deep learning framework for missing data imputation introduced in Khadem Charvadeh et al. (2025) <doi:10.1002/sim.70335>. The model accommodates all three types of missing data mechanisms: Missing Completely At Random (MCAR), Missing At Random (MAR), and Missing Not At Random (MNAR). While it is particularly well-suited to MNAR scenarios, where missingness patterns carry informative signals, 'CISS-VAE' also functions effectively under MAR assumptions.