Bayesian optimization using Gaussian Process
Bayesian Optimization using Gaussian Process to optimize an objective function with respect to input controls.
For the most basic usage, run basic_gp_example.py, which is an example code to run as-is without any input arguments. For an interactive version, check out the Jupyter notebook 'Plug_and_play-basic example.ipynb'
This tool was developed primarily for particle accelerators with various machine interfaces. It has the capability to run on:
- Real machine (see 'Pegasus_UCLA.py' example)
- Simulation (see 'Spear3_SLAC.py' example)
- Surrogate model of the machine (see 'APS_ANL.py' example)
- Analytic function (see 'basic_example.py')
Instructions to build your own optimizer
Step 1. Build your 'machine interface' file: This file will contain the class object that will represent your target function. To build it, open machine_interfaces/machine_interface_example.py and follow the step-by-step instructions in the comments at the top of the file.
Step 2. Build your 'scan_params.npy' file: This is the .npy file that will contain the settings for the optimizer to load and use. To build it, open modules/make_scan_params_file.py and follow the step-by-step instructions in the comments at the top of the file.
Step 3. Run the optimization: For the most basic usage, open basic_gp_example.py and follow the step-by-step instructions in the comments at the top of the file.