mlr3mbo: Bayesian Optimization in R

2025-06-13 · Source: stat.ML updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Advanced, extended

Summary

mlr3mbo is a new, comprehensive, and modular R toolbox for Bayesian optimization (BO), supporting single- and multi-objective optimization, multi-point proposals, batch and asynchronous parallelization, and robust error handling. It is designed for both applied settings and research, allowing users to construct custom BO algorithms from flexible building blocks. The software underwent extensive empirical evaluations on the YAHPO Gym benchmark suite, identifying robust default configurations for numeric and mixed-hierarchical optimization. Benchmarking against state-of-the-art optimizers like HEBO, SMAC3, Ax, and Optuna demonstrated that mlr3mbo achieves competitive performance, particularly excelling in runtime efficiency on numeric instances. The package integrates seamlessly with the mlr3 ecosystem for machine learning and hyperparameter optimization (HPO).

Key takeaway

For AI Engineers and Research Scientists engaged in black-box optimization or hyperparameter tuning in R, mlr3mbo provides a powerful and flexible solution. Its modular design allows for both standard, high-performance configurations and custom algorithm development. Consider adopting mlr3mbo for your next optimization task, especially if you require competitive performance, efficient runtimes, and robust error handling across diverse search spaces, including complex mixed-hierarchical structures.

Key insights

mlr3mbo offers a modular R framework for Bayesian optimization, achieving state-of-the-art performance with flexible configurations.

Principles

• Modular design enhances extensibility and customizability.
• Surrogate models and acquisition functions are key BO components.
• Error handling and transformations improve BO robustness.

Method

Bayesian optimization iteratively fits a probabilistic surrogate model to an archive of evaluations, then optimizes an acquisition function to propose new points, balancing exploration and exploitation.

In practice

• Use `regr.km` (GP) for numeric search spaces.
• Employ `regr.ranger` (RF) for mixed-hierarchical spaces.
• Apply logarithmic transformations for heavy-tailed objectives.

Topics

• mlr3mbo
• Bayesian Optimization
• R Programming Language
• Hyperparameter Optimization
• Surrogate Models

Code references

• mlr-org/mbo_config
• mlr-org/mlr3mbo

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Student

Related on AIssential

• In-Context Learning for Latent Space Bayesian Optimization — Adapting tabular foundation models for latent space Bayesian optimization requires specific pretraining to address domain mismatches.
• RL2ML: Finite-Rollout Surrogate Objectives from Reinforcement Learning to Maximum Likelihood — RL2ML offers unbiased, finite-rollout surrogate objectives connecting RL and ML, with optimal choice depending on metrics and variance.
• Typed Component Algebras for Simulated Annealing and Markov-Chain Monte Carlo — Decomposing SA/MCMC into a typed component algebra enables modular improvements and verifiable correctness.
• Model-Agnostic Meta Learning for Class Imbalance Adaptation — HAMR adaptively reweights and resamples data to improve NLP model performance on imbalanced and difficult classes.
• A Human-in-the-Loop Bayesian Optimization Framework for Constraint-Aware Bioprocess Development — This framework extends Human-in-the-Loop Bayesian Optimization for robust, constraint-aware bioprocess development.
• skwdro: a library for Wasserstein distributionally robust machine learning — skwdro simplifies robust ML model training using Wasserstein distributionally robust optimization with PyTorch and scikit-learn compatibility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.