mo-bayes-opt 0.1.0

Multi-Objective Bayesian Optimization with Gaussian Processes and adaptive noise

🎯 Multi-Objective Bayesian Optimization (MOBO)

A modular Python library for multi-objective Bayesian optimization using Gaussian Processes (GP) and BoTorch, with optional support for adaptive noise modeling.

Built with PyTorch, GPyTorch, and BoTorch

---

📦 Features

• ✔️ Multi-objective optimization with EHVI / ALT-EI
• ✔️ Independent GP models per objective
• ✔️ Adaptive noise learning using a neural NoiseNet
• ✔️ Pareto front extraction and visualization
• ✔️ Hypervolume logging per step
• ✔️ Easy-to-extend architecture for experiments

---

📁 Project Structure

mo_bayes_opt/ ├── models/ # GPModel, NoiseNet, GPTrainer ├── acquisition/ # Acquisition function optimization ├── core/ # BO loop & objective functions ├── utils/ # Logging, visualization ├── experiments/ # Example experiments ├── data/ # Hypervolume log storage

---

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run Example Optimization

def joint_objective(x):
    x1 = x[:, 0]
    x2 = x[:, 1]
    interaction = torch.sin(5 * torch.pi * x1 * x2)
    f1 = ((x[:, :3] * torch.sin(3 * torch.pi * x[:, :3])).sum(dim=-1) + 0.5 * interaction)
    f2 = (((1 - x[:, 3:]) * torch.cos(3 * torch.pi * x[:, 3:])).sum(dim=-1) - 0.3 * interaction)
    return torch.stack([f1, f2], dim=-1)

bo = MultiObjectiveBO(
    objective_fn=joint_objective,
    input_dim=3,
    bounds=torch.tensor([[0.0] * 3, [1.0] * 3]),
    ref_point=torch.tensor([0.0, 0.0]),
    strategy="EHVI",
    use_adaptive_noise=False
)
hypervolumes = bo.run(num_repeats=1, num_queries=100)

3. Output

Console: Logs trial progress and hypervolume

File: data/hypervolume_log.csv containing hypervolume progression

📈 Visualization

Use plot_pareto(train_y, trial, step) to visualize the Pareto front at a given step.