Surrogate Learning in Meta-Black-Box Optimization: A Preliminary Study

Here we provide source codes of Surr-RLDE, which has been recently accepted by GECCO 2025.

Citation

The PDF version of the paper is available here. If you find our Surr-RLDE useful, please cite it in your publications or projects.

@inproceedings{ma2025surrogate,
  title={Surrogate learning in meta-black-box optimization: A preliminary study},
  author={Ma, Zeyuan and Huang, Zhiyang and Chen, Jiacheng and Cao, Zhiguang and Gong, Yue-Jiao},
  booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},
  pages={1137--1145},
  year={2025}
}

Requirements

You can install all of dependencies of Surr-RLDE via the command below.

pip install -r requirements.txt

Train

Surrogate Learning Stage

The surrogate learning process can be activated via te command below

python main.py --train_surrogate 

The trained model will be saved tooutput/surrogate_model/

Policy Learning Stage

The Surr-RLDE agent training process can be activated via the command below, which is just an example.

python main.py --run_experiment --problem bbob-surrogate 

For more adjustable settings, please refer to main.py and config.py for details.

Recording results: Log files will be saved to ./output/train/ . The saved checkpoints will be saved to ./agent_model/train/. The file structure is as follow:

|--agent_model
   |--train
      |--Surr_RLDE_Agent
         |--run_Name
            |--checkpoint0.pkl
            |--checkpoint1.pkl
            |--...

|--output
   |--train
      |--Surr_RLDE_Agent
         |--runName
            |--log
            |--pic

Test

The test process can be easily activated via the command below. The defalt agent load path is agent_model/test/

python main.py --test --agent_load_dir YourAgentDir --agent_for_cp Surr_RLDE_Agent --l_optimizer_for_cp Surr_RLDE_Optimizer 

You can compare Surr-RLDE with DEDQN, DEDDQN, GLEET by adding them into the agent_for_cp and l_optimizer_for_cp

python main.py --test --agent_load_dir YourAgentDir --agent_for_cp Surr_RLDE_Agent DEDQN_Agent --l_optimizer_for_cp Surr_RLDE_Optimizer DEDQN_optimizer

We highly recommend you visit MetaBox 2.0. In this repository, you will find additional baselines and improved parameter settings for the Surr-RLDE, as well as an expanded set of BBOB-surrogate problems.