The JAMPOT Project

The goal of JAMPOT (JAva Multifidelity and Probabilistic Optimization Toolbox) is to provide a flexible, platform-independent API that can be used for implementing and testing algorithms for simulation-based optimization.

See:
Description

Packages

caffeinterface	Provides classes to interface with CAFFE classes, and convert them to ScalarFunctions and VectorFunctions.
data	Provides classes for handling sets of double-valued input-output data.
distribution	Provides classes to handle multivariate continuous distributions.
diviner	Provides functionality for measuring the level of promise of a design candidates.
experiment
function	Provides classes for scalar and vector-valued functions that operate on multi- dimensional double-valued data.
iterator	Provides iterator classes.
model	Provides classes that handle various kinds of approximation models, data fits, and so on.
problem	Provides classes to create, modify and use various kinds of optimization problems using objective functions and inequality constraints.
protobuf
util	Provides various utility classes.

The goal of JAMPOT (JAva Multifidelity and Probabilistic Optimization Toolbox) is to provide a flexible, platform-independent API that can be used for implementing and testing algorithms for simulation-based optimization. This means that objective and constraints are evaluated by a computationally expensive simulation, and that derivatives are usually not available. The focus of this toolbox is the use of multifidelity techniques that allow the integration of two (or more, in the future) analyses of varying computational cost and accuracy, for all of the simulations involved in the optimization problem.

Core Features

• Platform independence.
• Uses colt libraries for efficient numerical computation.
• Uses the powerful Spring Framework for dependency injection, thus enabling ...
• Simple user-interface using xml files.
• Extensive documentation.

Specific Features

• Support for multivariate continuous probability distributions (This is surprisingly rare.)
• Straightforward method for specifying optimization problems.
• Robust Gaussian Process Regression implementation.
• Full support for shell commands from within Java, including complete output and error stream access (much more powerful than just Runtime).
• Includes (easily extensible) interfaces for legacy codes (such as A502).
• Coming soon: CaffeApp support for integration with PASS.