Hybrid Tuner for Hyperparameter Tuning

Installations

Install the following directory with: 'pip install -e .'

Dependencies

HybridTuner is usable with only Python libraries.

However, HybridTuner greatly benefits from the use of:

1. MATLAB
2. TOMLAB glcDIRECT

All of the results reported in the corresponding paper make use of the TOMLAB and MATLAB solvers described within.

Examples

We have included several examples in the Example directory. To run an example file, modify 'matlab_path' to the location of your MATLAB installation, or remove it if you do not have MATLAB installed.

To tune your own application, modify the myexec file to call your black-box function.

Required Format:

1. Parameters are read from myin into black-box function
2. Output is written to myout
3. Parameter lower and upper bounds are defined in the file myparams.json

Example outputs: when using MATLAB and solver = 8:

Executing "python example.py myparams.json" in the ./examples/BanditDFO directory will return the following:

"Bandit DFO has completed"

"Best Solution = 1.0 found after 105 iterations"

Papers

• B. Sauk and N.V. Sahindis. HybridTuner: Tuning with hybrid derivative-free optimization initialization strategies.

Support

• This work was conducted as part of the Institute for the Design of Advanced Energy Systems(IDAES) with funding from the Office of Fossil Energy, Cross-Cutting Research, U.S. Department of Energy.
• This work used the Extreme Science and Engineering Discovery Environment (XSEDE),which is supported by National Science Foundation grant number ACI-1548562. Specifically, it used the Bridges system, which is supported by NSF award number ACI-1445606, at the Pittsburgh Super-computing Center (PSC).
• We also gratefully acknowledge the support of the NVIDIA Corporation with the donation of the NVIDIA Tesla K40 GPU used for this research.