Bayesian optimization in high-dimensions via random embedding.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
demos First commit Aug 4, 2013
src First commit Aug 4, 2013
README.md First commit Aug 4, 2013
startup.m First commit Aug 4, 2013

README.md

REMBO

This package contains code for the paper "Bayesian Optimization in a Billion Dimensions via Random Embeddings". The paper tries to solve the problem of doing Bayesian optimization in high dimensions. For more details please read the paper (http://www.cs.ubc.ca/~ziyuw/papers/rembo.pdf).

INSTALL

The package is written in Matlab. Thus to install, just run startup.m. This package could make use of the function "fminsearch" from Matlab. This, however, is not necessary.

To run the lpsolve example from the paper, please download the data files from: http://www.cs.ubc.ca/~ziyuw/AC_blackbox_eval.tar. And untar the data files to the folder /demos/lpsolve/.

Configuration files

This package allows one the specify the parameters settings in a configuration file. In the configuration file, one can specify the type of parameters to be optimized. Currently, the package supports continuous, discrete, and categorical parameters. It also allows one to use log scales. For an example of such a configuration file, please checkout "demos/lpsolve/lpsolve.yaml".