Skip to content

examples

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

examples

README.md

mpBAX Examples

This directory contains examples demonstrating different mpBAX features and workflows.

Two Approaches

mpBAX supports two complementary workflows:

Python API (Examples 01-07): Pass classes and functions directly in config dict. Best for rapid prototyping and simple scripts.

YAML API (yaml_example/): Specify import paths in YAML file. Best for production, configuration management, and team collaboration.

Setup

Option 1: pip install (Recommended)

cd /path/to/mpBAX
pip install -e .

Then run any example:

python examples/01_basic_optimization.py
cd examples/yaml_example && python run.py

Option 2: PYTHONPATH (No installation)

export PYTHONPATH=/path/to/mpBAX:$PYTHONPATH
python examples/01_basic_optimization.py

Example Overview

yaml_example/

YAML configuration workflow - Recommended for production!

  • Complete experiment in YAML config file
  • Oracle functions in separate importable module
  • Ideal for configuration management and reproducibility
  • See yaml_example/README.md for details

01_basic_optimization.py

The simplest example - Start here!

  • Single oracle optimization
  • Quadratic function minimization
  • DummyModel + GreedySampling
  • ~60 lines total

02_multi_oracle.py

Multiple independent oracles

  • Two oracles: Sphere and Rosenbrock functions
  • Each oracle has its own model and data
  • Single algorithm proposes for both
  • Useful for multi-objective optimization

03_multi_output.py

Vector-valued objectives

  • Oracle returns multiple outputs (shape (n, k) with k>1)
  • Model predicts all outputs simultaneously
  • Useful for multi-task learning scenarios

04_checkpointing.py

Saving and resuming

  • Automatic checkpointing every loop
  • Resume from latest checkpoint
  • Resume from specific loop number
  • Essential for long-running optimizations

05_custom_model.py

Extending the framework

  • Define custom BaseModel subclass
  • Implement train() and predict() methods
  • Example: Simple GP-like model with RBF kernel
  • Shows how easy it is to add new models

06_danet_model.py

Deep learning plugin (requires PyTorch)

  • DANetModel plugin for neural network surrogate
  • Demonstrates finetune mode
  • Sample weighting (recent data emphasized)
  • Adaptive epochs (100 initial, 20 incremental)

07_flexible_config.py

New flexible config patterns (mpBAX v2)

  • Demonstrates all new parameter placement options
  • input_dim in model.params instead of oracle level
  • Default generator shortcut (generate.params.n)
  • Custom generator with all params (n, d, custom)
  • 'training' instead of 'model' for top-level config
  • Shows backward compatibility with traditional patterns

08_checkpoint_analysis.py

Post-optimization analysis utilities

  • Load models from specific loops
  • Get fn_pred_list for testing with algorithms
  • Load data from individual loops or accumulated
  • Examine optimization progress and convergence
  • Test historical models with current algorithms
  • Perfect for notebooks and analysis workflows

What to Try Next

After running the examples, try:

  1. Modify parameters: Change n_initial, max_loops, n_propose
  2. Change oracle functions: Try different test functions
  3. Experiment with models: Swap DummyModel for your custom model
  4. Add visualization: Plot optimization progress
  5. Real applications: Replace toy functions with your actual simulations

Need Help?

  • See main README.md for API reference
  • Check tests/ for more code examples
  • See mpbax/plugins/models/README.md for DANetModel details