This is an implementation of Bayesian Optimization for Cascade-type Multistage Processes (Kusakawa et al., 2022).
This page provides instructions, shell scripts, python codes, and binary files to reproduce the experimental results.
- Hardware platform: x86_64
- OS: CentOS Linux release 7.7.1908 (Core) / Ubuntu 20.04.2 LTS
- Python version: 3.9.6
- Required python packages: listed in
requirements.txt
.- Run
pip install -r requirements.txt
to install these packages.
- Run
-
Download or clone this repository and make sure that
emulator/
,scripts/
, andsrc/
are located as follows:./ # project directory ├── emulator/ │ ├── 64bit/ │ └── ECdata/ ├── scripts/ └── src/
-
To reproduce the experimental results, run the following commands. When each shell script finishes, a PDF file in which the experimental results are plotted will be created in
./figure
(./figure
directory will be created automatically). If an argument of bash file is nothing, experiments are performed sequentially. The experiments can be parallelized by ``bash ./scripts/figure3_sample_path_3.sh parallel_num'' in all the scripts.-
Figure 3: Synthetic functions
bash ./scripts/figure3_sample_path_3.sh bash ./scripts/figure3_sample_path_5.sh bash ./scripts/figure3_rosenbrock_3.sh bash ./scripts/figure3_rosenbrock_5.sh bash ./scripts/figure3_sphere.sh bash ./scripts/figure3_matyas.sh
-
Figure 4: Solar cell simulator
bash ./scripts/figure4_solarcell.sh
-
Figure 5: Hydrogen plasma treatment process
bash ./scripts/figure5_hydrogen_plasma.sh
-
Figure 6: Suspension setting (sample path, solar cell simulator)
bash ./scripts/figure6_sample_path_3.sh bash ./scripts/figure6_sample_path_5.sh bash ./scripts/figure6_solarcell.sh
-
Figure 7: Comparison between EI-based and EI-FN
bash ./scripts/figure7_sample_path_3.sh bash ./scripts/figure7_sample_path_5.sh
-
S. Kusakawa, S. Takeno, Y. Inatsu, K. Kutsukake, S. Iwazaki, T. Nakano, T. Ujihara, M. Karasuyama, I. Takeuchi, Bayesian Optimization for Cascade-type Multi-stage Processes, Neural Computation, 2022. (to appear)
Preprint Paper: https://arxiv.org/abs/2111.08330