Package {smriti}

FIML-Smriti Refinement Wrapper

Description

This function serves as an agnostic refinement layer, combining Full Information Maximum Likelihood (FIML) with the structural covariance preservation of smriti_impute().

Usage

smriti_fiml(data, model, initial_imputation = NULL, lambda = 1)

Arguments

Value

A data frame with FIML-consistent, covariance-projected imputed values.

missForest-Smriti Refinement Wrapper

Description

This function serves as an agnostic refinement layer, combining the predictive power of Random Forests via missForest with the structural covariance preservation of smriti_impute().

Usage

smriti_forest(data, time_cols, robust = TRUE, ...)

Arguments

Value

A data frame with covariance-projected imputed values.

Examples

## Not run: 
df <- data.frame(T1 = c(1, NA, 3, 4), T2 = c(NA, 2, 3, 4))
smriti_forest(df, time_cols = 1:2)

## End(Not run)

Smriti Automated Longitudinal Imputation

Description

This function performs an automated routing and refinement for longitudinal missing data. It establishes a target covariance manifold from observed data, and then projects an initial imputation back toward the structural manifold using a Lagrangian constraint. Only originally-missing values are updated; observed data is held fixed.

Usage

smriti_impute(
  data,
  time_cols,
  initial_imputation = NULL,
  lambda = 1,
  learning_rate = 0.001,
  tol = 1e-06,
  max_iter = 2000,
  robust = FALSE,
  custom_target = NULL
)

Arguments

Value

A data frame with imputed and structurally refined values. Only the originally-missing cells are modified; observed values are returned unchanged.

Examples

# Simulated longitudinal data with scattered missingness
df <- data.frame(
  T1 = c(1.2, NA, 2.8, 3.1),
  T2 = c(2.1, 2.5, NA, 4.0),
  T3 = c(3.0, 3.3, 4.1, NA)
)
smriti_impute(df, time_cols = 1:3, robust = FALSE)

Smriti Multiple Imputation Wrapper

Description

This function generates multiple imputed datasets to capture the uncertainty inherent in missing data imputation. It uses bootstrapping to create distinct datasets, applies the Smriti Lagrangian refinement to each, and returns a list of completed datasets suitable for pooling.

Usage

smriti_mi(data, time_cols, m = 5, initial_imputation = NULL, ...)

Arguments

Details

Approximate (bootstrap) multiple imputation. Proper multiple imputation (Rubin 1987) draws imputations from the Bayesian posterior predictive distribution of the missing data given the observed data and the imputation model. This function instead bootstraps the rows and then deterministically applies smriti_impute() to each bootstrap sample. The between-imputation variance therefore captures sampling variability but not the full imputation-model uncertainty. The resulting standard errors may be moderately anti-conservative. Users needing valid Rubin's-rules pooling should treat the output as approximate MI and consider adding a stochastic residual-draw step, or use a fully Bayesian engine such as mice.

Pooling the returned list can be performed manually via Rubin's rules:

mi_list <- smriti_mi(df, time_cols = 2:5, m = 10)
# Within-imputation estimates
estimates <- lapply(mi_list, function(d) coef(lm(y ~ x, data = d)))
# Pool with Rubin's rules (point estimate = mean, SE = sqrt(W + (1+1/m)*B))

Value

A list of m completed data frames, each with an "imputation" attribute giving its index (1..m). The list has class "smriti_mi_list".

See Also

smriti_impute() for the single-imputation engine.

Examples

## Not run: 
df <- data.frame(
  T1 = c(1, 2, NA, 4),
  T2 = c(NA, 2, 3, 4),
  T3 = c(1, NA, 3, 4)
)
mi_list <- smriti_mi(df, time_cols = c("T1", "T2", "T3"), m = 5)

## End(Not run)

mice-Smriti Refinement Wrapper

Description

This function serves as an agnostic refinement layer, combining multivariate imputation by chained equations (MICE) with the structural covariance preservation of smriti_impute().

Usage

smriti_mice(data, time_cols, robust = TRUE, ...)

Arguments

Value

A data frame with covariance-projected imputed values.

Examples

## Not run: 
df <- data.frame(T1 = c(1, NA, 3, 4), T2 = c(NA, 2, 3, 4))
smriti_mice(df, time_cols = 1:2)

## End(Not run)

missRanger-Smriti Refinement Wrapper

Description

This function serves as an agnostic refinement layer, combining the fast predictive imputation of missRanger with the structural covariance preservation of smriti_impute().

Usage

smriti_ranger(data, time_cols, robust = TRUE, ...)

Arguments

Value

A data frame with covariance-projected imputed values.

Examples

## Not run: 
df <- data.frame(T1 = c(1, NA, 3, 4), T2 = c(NA, 2, 3, 4))
smriti_ranger(df, time_cols = 1:2)

## End(Not run)