skills.rest
learningmatter-mit
learningmatter-mitOFFICIAL
Explore all skills by this author →
WebsiteGitHub

ml-bayesian-optimization

Optimize expensive black-box objectives via Gaussian-process surrogate and Expected Improvement.

144|21|Updated Jan 8, 2026
Search Tags:#hyperparameter tuning#bayesian optimization#multi-objective#gaussian process#materials discovery#candidate suggestions
One-click install
npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-bayesian-optimization
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ml-bayesian-optimization
Source: https://github.com/learningmatter-mit/AtomisticSkills/tree/main/.agents/skills/ml-bayesian-optimization
Command: npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-bayesian-optimization

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires scikit-learn, scipy, numpy, pandas, pyyaml, matplotlib, and includes scripts (resource) components.

What problem does it solve?

This Skill solves the problem of efficiently optimizing expensive black-box objectives (like simulation outputs or materials properties) while minimizing the number of costly evaluations required.

Core Features & Use Cases

  • Guides iterative experiments or simulations: Builds a surrogate model over evaluated data and proposes the next most promising candidates using Bayesian Optimization.
  • Supports single- and multi-objective optimization: Uses Expected Improvement for single-objective and ParEGO-style scalarization for multi-objective campaigns.
  • Works with MCP-backed evaluators: Outputs candidate parameter sets that you can evaluate with MCP tools (e.g., relaxation, bandgap prediction, DFT workflows) and then feed back into the next BO round.
  • Includes analysis and visualization: Produces convergence plots, parameter importance, Pareto front (for 2 objectives), and GP model visualizations (for 1–2 range parameters).

Quick Start

Run a BO initialization campaign by generating Sobol-sampled candidates from your search space with an output CSV, then evaluate those candidates externally and append results to evaluated.csv for the next BO round.

Frequently Asked Questions about ml-bayesian-optimization

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I optimize expensive black-box simulation functions with minimal evaluations?▼

Bayesian optimization efficiently optimizes expensive black-box objectives by iteratively learning a probabilistic Gaussian-process surrogate and proposing informative next candidates to minimize costly evaluations.

What is the best way to perform multi-objective hyperparameter tuning for materials discovery?▼

Multi-objective optimization uses ParEGO-style scalarization to balance competing objectives, providing Pareto front visualizations and parameter importance for materials or chemistry design loops.

How do I define a search space for Bayesian optimization using scikit-learn and scipy?▼

Define your search space in a YAML file, run a Sobol-sampled initialization campaign to generate candidates, evaluate them externally, and append results to a CSV for the next round.

Does this Bayesian optimization approach work with external MCP evaluators and DFT workflows?▼

Yes, it outputs candidate parameter sets as CSV files for external evaluation with MCP-backed tools like DFT workflows or bandgap prediction, then feeds results back into the next optimization round.

Can I visualize convergence plots and Gaussian process models for single-objective optimization?▼

Yes, the analysis produces convergence plots, parameter importance rankings, and GP model visualizations for one to two range parameters to track optimization progress.

When should I not use Gaussian process surrogate models for hyperparameter tuning?▼

Gaussian process surrogates with Matern kernels are not ideal when evaluation costs are trivially low or when search spaces exceed a few dozen dimensions, as GP training scales poorly with data size.

Related Skills

Explore More Education & Research Skills→
PharMolixPharMolix
1.1k

mutation-design-gfp

Design GFP mutant libraries through iterative optimization and export 96-sequence CSV files.

Community
Advanced
learningmatter-mitlearningmatter-mit
144

mat-dft-electron-phonon

Compute temperature-dependent bandgap renormalization via atomate2 DFT electron-phonon workflows.

Official
Advanced
nahisahonahisaho
3

scientific-process-optimization

Optimize process parameters using ML-RSM and Pareto fronts.

Community
Advanced
snoodleboot-iosnoodleboot-io
2

hyperparameter-optimization

Explain hyperparameter tuning techniques including grid search, random search, Bayesian optimization, and reinforcement learning.

Official
Intermediate