• This repository contains a Dockerized Flask application that serves a benchmark for use within expensive, black-box, and multi-fidelity optimization frameworks.
• The Flask application can be accessed via a REST API and supports both single and multi-fidelity evaluations with adjustable CPU usage for parallel simulations.

0. Install Docker

1. Clone the repository.

   git clone https://github.com/trsav/reactor_benchmark.git
   cd reactor_benchmark

2. Build the Docker image. This will take approximately 10 minutes

   docker build -t benchmark .

   If this step fails (i.e. ERROR: error getting credentials - err: exit status 1, out: ) you may need to log into Docker first using $ docker login, or play around with using sudo.

3. Run the Docker image.

   docker run -p 5001:5001 benchmark

4. Send a POST request to the Flask application (reactor_design_problem/test_eval.py).

   import requests
   import json
   import numpy as np
   
   url = "http://localhost:5001/cross_section"
   d = {"x": list(np.random.uniform(0, 1, 36)), "z": [0.5, 0.5], "keep_files": False, "cpus": 2}
   headers = {"Content-Type": "application/json"}
   response = requests.post(url, headers=headers, data=json.dumps(d))
   print(response.text)

Refer to the function description in the code for more information about the x, z, keep_files, and cpus parameters.

• To perform single fidelity evaluations either omit z from the POST dictionary (in which case the simulation will be performed with fidelities [0.75,0.75]) or choose your own values and keep track of them!
• Lower fidelities will be quicker to evaluate, but will probably provide simpler, more flat objective functions.
• A property of this problem is that lower fidelities are significantly biased from higher fidelities, providing consistently higher objective values.

• Docker (if using Windows, you will need to install WSL)
• This software has been tested on a 2019 Macbook Pro, and a Windows PC.

• Flexible: Supports both single and multi-fidelity evaluations.
• Parallelizable: Can adjust CPU usage for parallel simulations.
• Easy to use: Accessible via a REST API.

This project is licensed under the terms of the MIT license.