smriti

smriti is an R package for automated longitudinal missing data imputation. It combines the predictive flexibility of non-parametric machine learning with a C++ Lagrangian projection engine to strictly preserve the structural variance of the target covariance manifold.

Installation

# Stable CRAN release
install.packages("smriti")

# Development version
# install.packages("devtools")
devtools::install_github("xguot/smriti")

Usage

Impute longitudinal missing data while preserving the underlying covariance structure:

library(smriti)

imputed_data <- smriti_impute(
  data = clinical_df, 
  time_cols = c("V1", "V2", "V3", "V4"),
  lambda = 0.5,
  robust = TRUE  # Enables the MCD estimator to suppress outliers
)

Architecture

The imputation pipeline executes in three phases:

1. Initialization: Generates a dense preliminary point-cloud via Random Forest (missForest).
2. Manifold Mapping: Establishes the target covariance structure from observed data, with optional robust estimation.
3. Lagrangian Routing: Projects the initial matrix back onto the structural manifold via a constrained gradient descent update.

Citation

If you utilize smriti in your research, please cite:

Guo, X. (2026). smriti: Structural Variance Preservation for Longitudinal Missing Data Imputation. R package version 0.1.0.