Professional AssociationsProfessional Associations: DIMACS | INFORMS | Mathematical Optimization Society
I write a lot of software for my research. Part of it can be downloaded from this page, if there are no licensing issues. Some of my other contributions can be found in larger projects such as COIN-OR Couenne. This software is provided "as is", without any warranty. Most of this software was written before joining IBM.
RBFOpt is a library for black box (derivative free) optimization of functions with costly function evaluation. For more information, look at the paper RBFOpt. The library is released under the Revised BSD license.
RBFOpt main web page on COIN-OR and on developerWorks Open. The software is available on GitHub and on PyPI.
Testing the safety of cut generators
This project provides a framework for testing and comparing the safety of tableau cut generators. It is released under GNU GPL v3. The archive contains the source code (in C, C++, Python), data sets, scripts, as well as detailed instructions. For more information, please refer to the paper On the safety of Gomory cut generators.
Download source (from Francois Margot's webpage)
Toolbox for approximate stochastic DP
This is a Fully Polynomial-Time Approximation Scheme for convex stochastic dynamic programs. A class of stochastic dynamic programs that satisfy certain convexity requirements falls into this framework, but the code can be used heuristically on other types of dynamic programs as well. The code is released under GNU GPL v3. The archive contains the source code (in Python 3). I provide both an exact dynamic programming algorithm and an FPTAS for this class of problems. The code can be run using floating point arithmetics. An implementation that uses infinite precision arithmetics is available but not supported. For more information, please refer to the paper A computationally efficient FPTAS for convex stochastic dynamic programs. Since there are legal issues hosting code here on IBM's wepage, this archive is only available upon request.
RECIPE is a primal heuristic for nonconvex MINLPs. It has global and local search phases, and it can use any solver with an AMPL interface as local subsolvers. The global search phase is coordinated by a Variable Neighborhood Search. The code is distributed under GNU GPL v3, only for the parts that we developed. (For instance, we use AMPL Netlib, for which I could not determine the license). The archive contains the source code (in C, AMPL, bash), some scripts for testing, and detailed instructions. For more information, please refer to the paper A recipe for finding good solutions to MINLPs.
Download source (from Leo Liberti's webpage).