miceFast

Author: Maciej Nasinski

Check the miceFast website for more details

Overview

miceFast provides fast methods for imputing missing data, leveraging an object-oriented programming paradigm and optimized linear algebra routines.
The package includes convenient helper functions compatible with data.table, dplyr, and other popular R packages.

Major speed improvements occur when:
- Using a grouping variable, where the data is automatically sorted by group, significantly reducing computation time. - Performing multiple imputations, by evaluating the underlying quantitative model only once for multiple draws. - Running Predictive Mean Matching (PMM), thanks to presorting and binary search.

For performance details, see performance_validity.R in the extdata folder.

It is recommended to read the Advanced Usage Vignette.

Installation

You can install miceFast from CRAN:

install.packages("miceFast")

Or install the development version from GitHub:

# install.packages("devtools")
devtools::install_github("polkas/miceFast")

Quick Example

Below is a short demonstration. See the vignette for advanced usage and best practices.

library(miceFast)

set.seed(1234)
data(air_miss)

# Visualize the NA structure
upset_NA(air_miss, 6)

# Simple and naive fill
imputed_data <- naive_fill_NA(air_miss)

# Compare with other packages:
# Hmisc
library(Hmisc)
data.frame(Map(function(x) Hmisc::impute(x, "random"), air_miss))

# mice
library(mice)
mice::complete(mice::mice(air_miss, printFlag = FALSE))

Loop example

Multiple imputations are performed in a loop where a continuous variable is imputed using a Bayesian linear model (lm_bayes) that incorporates relevant predictors and weights for robust estimation. Simultaneously, a categorical variable is imputed using linear discriminant analysis (LDA) augmented with a randomly generated ridge penalty.

library(dplyr)

# Define a function that performs the imputation on the dataset
impute_data <- function(data) {
  data %>%
    mutate(
      # Impute the continuous variable using lm_bayes
      Solar_R_imp = fill_NA(
        x = .,
        model = "lm_bayes",
        posit_y = "Solar.R",
        posit_x = c("Wind", "Temp", "Intercept"),
        w = weights  # assuming 'weights' is a column in data
      ),
      # Impute the categorical variable using lda with a random ridge parameter
      Ozone_chac_imp = fill_NA(
        x = .,
        model = "lda",
        posit_y = "Ozone_chac",
        posit_x = c("Wind", "Temp"),
        ridge = runif(1, 0, 50)
      )
    )
}

# Set seed for reproducibility
set.seed(123456)

# Run the imputation process 3 times using replicate()
# This returns a list of imputed datasets.
res <- replicate(n = 3, expr = impute_data(air_miss), simplify = FALSE)

# Check results: Calculate the mean of the imputed Solar.R values in each dataset
means_imputed <- lapply(res, function(x) mean(x$Solar_R_imp, na.rm = TRUE))
print(means_imputed)

# Check results: Tabulate the imputed categorical variable for each dataset
tables_imputed <- lapply(res, function(x) table(x$Ozone_chac_imp))
print(tables_imputed)

---

Key Features

• Object-Oriented Interface via miceFast objects (Rcpp modules).
• Convenient Helpers:
  • fill_NA(): Single imputation (lda, lm_pred, lm_bayes, lm_noise).
    
  • fill_NA_N(): Multiple imputations (pmm, lm_bayes, lm_noise).
    
  • VIF(): Variance Inflation Factor calculations.
    
  • naive_fill_NA(): Automatic naive imputations.
    
  • compare_imp(): Compare original vs. imputed values.
    
  • upset_NA(): Visualize NA structure using UpSetR.

Quick Reference Table:

Function	Description
new(miceFast)	Creates an OOP instance with numerous imputation methods (see the vignette).
fill_NA()	Single imputation: lda, lm_pred, lm_bayes, lm_noise.
fill_NA_N()	Multiple imputations (N repeats): pmm, lm_bayes, lm_noise.
VIF()	Computes Variance Inflation Factors.
naive_fill_NA()	Performs automatic, naive imputations.
compare_imp()	Compares imputations vs. original data.
upset_NA()	Visualizes NA structure using an UpSet plot.

---

Performance Highlights

Benchmark testing (on R 4.4.3, macOS M3 Pro, optimized BLAS and LAPACK) shows miceFast can significantly reduce computation time, especially in these scenarios:

• Linear Discriminant Analysis (LDA): ~5x faster.
  
• Grouping Variable Imputations: ~10x faster (and can exceed 100x in some edge cases).
  
• Multiple Imputations: ~x * (number of multiple imputations) faster, since the model is computed only once.
  
• Variance Inflation Factors (VIF): ~5x faster, because we only compute the inverse of X’X.
  
• Predictive Mean Matching (PMM): ~3x faster, thanks to presorting and binary search.

For performance details, see performance_validity.R in the extdata folder.