Software
Statistical computing, R packages and tools for reproducible research.
R and Python software I develop for statistical modelling and computation. I use C++ and RcppArmadillo in R packages, and Python for data processing, machine learning and reproducible research workflows.
R packages
High-dimensional regression, spatial and functional data analysis, shrinkage methods and dose–response models.
softridge
Soft-Thresholded Ridge Estimation
Soft-thresholded ridge estimation for high-dimensional linear regression. Computes fixed-tuning Gaussian coefficient risk for the estimable projection and prediction SURE, with cross-validation and least-squares refitting.
Cite this package
Arashi, M. and Yuzbasi, B. (2026). softridge: High-Dimensional Ridge Shrinkage and Selection by Soft Thresholding. R package version 0.3.0.1.
scaleCorrection
Optimism-Corrected Ridge Scaling
A research companion to softridge. Applies optimism-corrected scaling after soft thresholding and tunes SSTR alongside a matched Ridge-c0 control on a shared cross-validation surface. The correction is evaluated empirically.
Cite this package
Arashi, M. and Yuzbasi, B. (2026). scaleCorrection: Research Companion for Optimism-Corrected Ridge Scaling. R package version 0.1.0.
emaxProjection
Confidence-Region Projection for Integrated Emax Responses
Confidence-region projection for an integrated Emax response on a specified dose axis. Includes flat, linear and immediate-saturation limits, and reports outer intervals, feasible witnesses, endpoint gaps and termination status.
Cite this package
Yüzbaşı, B. (2026). emaxProjection: Confidence-Region Projection for Integrated Emax Responses. R package version 0.1.0.
gwrs
Fast Geographically Weighted Regression and Selection
Joint selection of predictor subsets, kernels and bandwidths in geographically weighted regression, with local Ridge, Lasso, Elastic Net, SCAD and MCP regularization. Includes spatial cross-validation and prediction at new locations.
AdaReFun
Adaptive Regional Functional Regression
Adaptive regional sparsity for scalar-on-function regression. Provides single- and multi-resolution regional estimation with MCP and SCAD penalties, model selection using an independent validation sample, and fixed-resolution fMCP/fSCAD estimators.
sglasso
Scaled Group Lasso
Scaled group lasso for variable selection with groupwise collinearity. Supports Gaussian and binomial responses, regularization paths and cross-validation, with fold-local preprocessing for binomial models.
Related article: Collinear Groupwise Selection via Scaled Group Lasso
lambertReg
Sparse Gaussian Regression with the Lambert Penalty
Sparse Gaussian regression with the Lambert penalty. Provides regularization paths with a fixed penalty shape, training-only cross-validation, original-scale predictions and numerical diagnostics.
rbridge
Restricted Bridge Estimation
Bridge regression with linear coefficient restrictions. Provides coefficient paths, cross-validation, prediction and coefficient extraction for restricted and unrestricted linear models, with computation through RcppArmadillo.
install.packages("rbridge")Related article: Penalized Regression via the Restricted Bridge Estimator
LHWL
Lorentz Hyperbolic Weighted Learning
Local regression and penalized learning using Lorentz hyperbolic representations. Includes Gaussian and binomial models, tuning, diagnostics and coefficient maps; representations need not be geographic.
HDMaxShrink
Test-Guided Shrinkage for High-Dimensional Linear Models
Test-guided shrinkage estimators for linear models, including settings with more predictors than observations. Combines maximum-statistic calibration with high-dimensional computational tools.
BernoulliRuns
Exact Run-Count Distributions for Independent Bernoulli Trials
Computes exact distributions of the total run count for independent Bernoulli trials with possibly different success probabilities. Provides probability masses, cumulative probabilities, expectations and one-sided tails.
penGWBR
Penalized Geographically Weighted Beta Regression
Geographically weighted beta regression for proportion responses, with LASSO and adaptive LASSO local variable selection. Includes spatial cross-validation, coefficient maps and diagnostics. The source repository is private; access is arranged with the author.
Local documentation is ready; source repository is private.
Cite this package
Yuzbasi, B. (2026). penGWBR: Penalized Geographically Weighted Beta Regression. R package source version 1.4.1.
Python packages
Robust regression and statistical learning under skewness and heavy tails.
ShapeAdaptiveEN
Shape-Adaptive Robust Regression
Python software for shape-adaptive elastic net (SA-EN) estimation under skewness and heavy tails. Provides ShapeAdaptiveEN and ShapePenalizedEN classes, fit/predict interfaces, diagnostic tools and a small usage example.
Research workflow
| Area | Use |
|---|---|
| R | Statistical modelling, simulation, package development and visualisation |
| Python | Package development, robust regression, machine learning and reproducible analysis |
| C++ / RcppArmadillo | Computationally intensive methods and performance-critical components in R packages |
| SPSS | Applied data analysis in the social sciences |