Data-Driven Multitask Optimization Laboratory

📖 Overview

D²MTOLab (Data-Driven Multitask Optimization Laboratory) is a comprehensive Python platform for optimization research, featuring 110+ algorithms, 180+ benchmark problems, and powerful experiment tools for problem definition, algorithm development, and performance evaluation.

Whether you're working on expensive black-box optimization, multiobjective optimization, or complex multitask scenarios, D²MTOLab provides a flexible and extensible framework to accelerate your research and support real-world applications.

✨ Features

• 🚀 Comprehensive Algorithms - 110+ algorithms for expensive/inexpensive, single/multitask, single/multiobjective, unconstrained/constrained optimization
• 📊 Rich Problem Suite - 180+ benchmark problems and real-world applications
• 🤖 Data-Driven Optimization - Surrogate modelling for expensive optimization
• 🔧 Flexible Framework - Simple API and intuitive workflow for rapid prototyping
• 🔌 Fully Extensible - Easy to add custom algorithms and problems
• ⚡ Parallel Computing - Multi-core support for batch experiments
• 📈 Powerful Analysis Tools - Built-in visualization and statistical analysis
• 📝 Complete Documentation - Comprehensive documentation and API reference

🚀 Quick Start

Installation

pip install ddmtolab

Requirements:

• Python 3.10+
• PyTorch 2.5+ (supports CPU, GPU optional for acceleration)
• NumPy 2.0+, SciPy 1.15+, scikit-learn 1.7+
• Matplotlib 3.10+, Pandas 2.3+

For detailed installation instructions, see Installation Guide.

Basic Usage

import numpy as np
from ddmtolab.Methods.mtop import MTOP
from ddmtolab.Algorithms.MTSO.MTBO import MTBO

# Step 1: Define objective function
def forrester(x):
    """Forrester function: (6x-2)^2 * sin(12x-4)"""
    return (6 * x - 2) ** 2 * np.sin(12 * x - 4)

# Step 2: Create optimization problem
problem = MTOP()
problem.add_task(forrester, dim=1)

# Step 3: Run optimization
results = MTBO(problem).optimize()

# Step 4: Display results
print(f"Best solution: {results.best_decs}")
print(f"Best objective: {results.best_objs}")

# Step 5: Analyze and visualize
from ddmtolab.Methods.test_data_analysis import TestDataAnalyzer
TestDataAnalyzer().run()

Batch Experiment

from ddmtolab.Methods.batch_experiment import BatchExperiment
from ddmtolab.Methods.data_analysis import DataAnalyzer
from ddmtolab.Algorithms.STSO.BO import BO
from ddmtolab.Algorithms.MTSO.MTBO import MTBO
from ddmtolab.Problems.MTSO.cec17_mtso_10d import CEC17MTSO_10D

if __name__ == '__main__':
    # Step 1: Create batch experiment manager
    batch_exp = BatchExperiment(
        base_path='./Data',      # Data save path
        clear_folder=True        # Clear existing data
    )

    # Step 2: Add test problems
    prob = CEC17MTSO_10D()
    batch_exp.add_problem(prob.P1, 'P1')
    batch_exp.add_problem(prob.P2, 'P2')

    # Step 3: Add algorithms with parameters
    batch_exp.add_algorithm(BO, 'BO', n_initial=20, max_nfes=100)
    batch_exp.add_algorithm(MTBO, 'MTBO', n_initial=20, max_nfes=100)

    # Step 4: Run batch experiments
    batch_exp.run(n_runs=20, verbose=True, max_workers=8)

    # Step 5: Run data analysis
    analyzer = DataAnalyzer()
    results = analyzer.run()

Optimization Process Visualization

D²MTOLab provides built-in animation tools to visualize the optimization process:

from ddmtolab.Problems.MTSO.cec17_mtso_10d import CEC17MTSO_10D
from ddmtolab.Algorithms.STSO.BO import BO
from ddmtolab.Algorithms.MTSO.MTBO import MTBO
from ddmtolab.Methods.animation_generator import create_optimization_animation

# Define problem
problem = CEC17MTSO_10D().P1()

# Run algorithms
BO(problem, n_initial=20, max_nfes=100, name='BO').optimize()
MTBO(problem, n_initial=20, max_nfes=100, name='MTBO').optimize()

# Generate animation
animation = create_optimization_animation(max_nfes=100, merge=2, title='BO vs MTBO')

🎯 Key Components

Algorithms (110+)

Problems (180+)

Methods

• MTOP Class: Flexible problem definition supporting single/multitask and single/multiobjective
• Batch Experiments: Parallel execution framework for large-scale experiments
• Data Analysis: Statistical analysis (mean, std, ranking) and visualization tools
• Performance Metrics: IGD, GD, IGD+, HV, DeltaP, Spacing, Spread, FR, CV
• Animation Generator: Optimization process visualization
• Algorithm Utilities: Reusable components (initialization, selection, operators)

🖥️ GUI (Experimental)

An experimental desktop GUI built with DearPyGui lives in the ui/ directory. It is shipped in this repository only (not part of the PyPI package):

pip install -r ui/requirements.txt
python ui/main.py

📚 Documentation

• Installation Guide
• Algorithms Reference
• Problems Reference
• Methods Guide
• Demo Scripts
• API Reference

📄 Citation

If you use D²MTOLab in your research, please cite:

@software{ddmtolab2025,
  author = {Jiangtao Shen},
  title = {D$^2$MTOLab: A Python Platform for Data-Driven Multitask Optimization},
  year = {2025},
  url = {https://github.com/JiangtaoShen/DDMTOLab}
}

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

📧 Contact

• Author: Jiangtao Shen
• Email: j.shen5@exeter.ac.uk
• Documentation: https://jiangtaoshen.github.io/DDMTOLab/
• Issues: GitHub Issues

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

Made with ❤️ by Jiangtao Shen