OPAL—A Framework for Optimization of Algorithms

What is OPAL?

OPAL is a Python modeling language for algorithmic optimization. Most algorithms depend on parameters. Although changing the values of those parameters doesn't affect the correctness of the algorithm, it typically affects its performance, where performance is understood broadly. How can we best choose those parameter values so as to maximize a certain measure of performance?

OPAL is a framework that allows to easily declare algorithms and the parameters on which they depend along with representative test cases. It provides a convenient syntax to formulate the optimization problem to be solved. A black-box optimization solver takes care of the rest.

Requirements

Python version 2.6 or 2.7 (not tested with 3.x)
NOMAD

Install

Unzip the package or clone the git repository
Go to the source directory
Run python setup.py install

Testing

To run the following test, Numpy is required. Install it with pip install numpy.

Assuming your PYTHONPATH is set correctly, you should be able to do:

cd examples/fd
python fd_optimize.py

References

Optimization of Algorithms with OPAL
Templating and Automatic Code Generation for Performance with Python
Taking Advantage of Parallelism in Algorithmic Parameter Optimization
Algorithmic Parameter Optimization of the DFO Method with the OPAL Framework

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!