stReans

An R package implementing batch and streaming k-means clustering with core computations written in C via the R native API.

Overview

stReans provides two complementary approaches to k-means clustering:

• kmeans_batch() - Lloyd's algorithm for datasets that fit in memory, with convergence detection based on centroid shift
• kmeans_stream_new() - incremental online k-means for data streams, processing one observation at a time without storing the full dataset

Installation

The package is not on CRAN. Install directly from GitHub:

# install.packages("devtools")
devtools::install_github("Luki308/stReans")

Usage

Batch K-Means

library(stReans)

# generate three well-separated clusters
set.seed(42)
cluster1 <- matrix(rnorm(40, mean = 0,  sd = 0.5), nrow = 20)
cluster2 <- matrix(rnorm(40, mean = 5,  sd = 0.5), nrow = 20)
cluster3 <- matrix(rnorm(40, mean = 10, sd = 0.5), nrow = 20)
X <- rbind(cluster1, cluster2, cluster3)

# run batch k-means
result <- kmeans_batch(X, k = 3L, max_iter = 100L, tol = 1e-6)

result$centers      # cluster centroids
result$assignments  # cluster label per observation
result$iterations   # how many iterations ran

Visualising Results

plot(X[, 1], X[, 2],
     col  = result$assignments,
     pch  = 16,
     cex  = 1.5,
     xlab = "X1",
     ylab = "X2",
     main = "Batch K-Means Clustering")

points(result$centers[, 1], result$centers[, 2],
       col = 1:3, pch = 4, cex = 3, lwd = 3)

legend("topright",
       legend = paste("Cluster", 1:3),
       col    = 1:3,
       pch    = 16)

Implementation Details

C Backend

• src/utils.c - Euclidean distance, column-major ↔ row-major matrix conversion utilities
• src/kmeans_batch.c - Lloyd's algorithm
• src/kmeans_stream_new.c - incremental online k-means
• src/init.c - native routine registration via R_registerRoutines

Development

# install dependencies
install.packages(c("devtools", "testthat", "roxygen2", "pkgdown"))

# clone and build
git clone https://github.com/Luki308/stReans.git
cd stReans

# in R
devtools::install()
devtools::test()
devtools::check()

Testing

Unit tests are written with testthat and run automatically via GitHub Actions on Linux, macOS, and Windows:

• tests/testthat/test-batch.R - correctness, edge cases, input validation
• tests/testthat/test-utils.R - distance and layout conversion utilities

License

MIT © Łukasz Lepianka