Spearmint is a software package to perform Bayesian optimization. The Software is designed to automatically run experiments (thus the code name spearmint) in a manner that iteratively adjusts a number of parameters so as to minimize some objective in as few runs as possible.
IMPORTANT: Spearmint is under an Academic and Non-Commercial Research Use License. Before using spearmint please be aware of the license. If you do not qualify to use spearmint you can ask to obtain a license as detailed in the license or you can use the older open source code version (which is somewhat outdated) at https://github.com/JasperSnoek/spearmint.
####Relevant Publications
Spearmint implements a combination of the algorithms detailed in the following publications:
Practical Bayesian Optimization of Machine Learning Algorithms
Jasper Snoek, Hugo Larochelle and Ryan Prescott Adams
Advances in Neural Information Processing Systems, 2012
Multi-Task Bayesian Optimization
Kevin Swersky, Jasper Snoek and Ryan Prescott Adams
Advances in Neural Information Processing Systems, 2013
Input Warping for Bayesian Optimization of Non-stationary Functions
Jasper Snoek, Kevin Swersky, Richard Zemel and Ryan Prescott Adams
International Conference on Machine Learning, 2014
Bayesian Optimization and Semiparametric Models with Applications to Assistive Technology
Jasper Snoek, PhD Thesis, University of Toronto, 2013
Bayesian Optimization with Unknown Constraints
Michael Gelbart, Jasper Snoek and Ryan Prescott Adams
Uncertainty in Artificial Intelligence, 2014
####Setting up Spearmint
STEP 1: Installation
- Install python, numpy, scipy. For academic users, the anaconda distribution is great. Use numpy 1.8 or higher.
- Download/clone the spearmint code
- Install the spearmint package using pip:
pip install -e \</path/to/spearmint/root\>
(the -e means changes will be reflected automatically) - Download and install MongoDB: https://www.mongodb.org/
- Install the pymongo package using e.g., pip
pip install pymongo
or anacondaconda install pymongo
STEP 2: Setting up your experiment
- Create a callable objective function. See ../examples/branin/branin.py as an example
- Create a config file. There are 3 example config files in the ../examples directory.
STEP 3: Running spearmint
- Start up a MongoDB daemon instance:
mongod --fork --logpath <path/to/logfile\> --dbpath <path/to/dbfolder\>
- Run spearmint:
python main.py \</path/to/experiment/directory\>
STEP 4: Looking at your results
Spearmint will output results to standard out / standard err. You can also load the results from the database and manipulate them directly.