README.md

Multiobjective Tree-structured Parzen Estimator (MOTPE)

NOTE

Journal paper is available!

@article{ozaki2022multiobjective,
  title={Multiobjective Tree-Structured Parzen Estimator},
  author={Ozaki, Yoshihiko and Tanigaki, Yuki and Watanabe, Shuhei and Nomura, Masahiro and Onishi, Masaki},
  journal={Journal of Artificial Intelligence Research},
  volume={73},
  pages={1209--1250},
  year={2022}
}

This repository is no longer maintained.
MOTPE is available on Optuna.
Please use the Optuna's MOTPE implementation (TPESampler).

Dependencies

Please install the following Python packages.

• ConfigSpace
• matplotlib
• numpy
• optproblems
• pandas
• pyDOE2
• pygmo
• scipy

Usage

Run MOTPE with the default settings on WFG4.

python main.py

Several commandline options are available.

python main.py --benchmark_name WFG4 --num_objectives 2 --num_variables 9 --k 1 --num_max_evals 250 --num_initial_samples 98 --init_method lhs --num_candidates 24 --gamma 0.10 --seed 128

Code Contributors

Yoshihiko Ozaki

Shuhei Watanabe