OR 2018: International Conference on Operations Research in Brussels, Belgium
GAMS will be at the OR 2018 in Brussels, Belgium
We will be at the OR 2018, the International Conference on Operations Research in Brussles, Belgium.
Visit our booth for a chat - give us feedback or share your experiences with GAMS.
You can also attend on one of our speeches:
GAMS and High-Performance Computing
By Fred Fiand and Michael Bussieck, GAMS Software GmbH
Solving challenging optimization problems often involves algorithms that could exploit the massive parallel computation power of modern High-Performance Computing (HPC) Architectures. GAMS has several features that support different modes of parallelism for some time. Recently, HPC capabilities of GAMS have been extended significantly. A link to parallel interior solver PIPS-IPM for large-scale block structured Linear Programs has been implemented. This link requires the user to communicate the model's block structure by variable annotation to the solver. GAMS facilitates user friendly annotation capabilities on the language level. Experimental implementations of parallel model generation for certain LPs illustrate additional speedup potential. Furthermore, the novel GAMS Embedded Code Facility, which extends the connectivity of GAMS to Python, allows the implementation of HPC applicable decomposition methods. Via packages like mpi4py the Message Passing Interface (MPI) standard can be used to implement efficient communication between master and worker processes as for example occurring in Benders Decomposition Approaches.
Model Deployment in GAMS
By Lutz Westermann, GAMS Software GmbH
In many cases, using GAMS in the typical fashion - i.e. defining and solving models and evaluating the results within the given interfaces – presents a sufficient way to deploy optimization models. The underlying field of mathematical optimization, in which the focus is not so much on visualization as on the problem structure itself, has remained a kind of niche market to this day. In the large and very extensive segment of business analytics, however, intuitive deployment and visualization are essential. Since these two areas are increasingly overlapping and in the context of the ever-increasing use of the Internet, interest in alternative deployment methods is also growing in the field of mathematical optimization. In this talk we will show how deployment options of GAMS models can look like. As an example, we present a web interface which is based on an R package called "Shiny". We will show how a model that was written entirely in GAMS can be deployed with this WebUI on either a local machine or a remote server (e.g. to leverage parallel computing) in just a few steps. While data manipulation, scenario management and graphical evaluation of the optimization results can then be performed from within the WebUI, the model itself is not changed. Therefore, the Operations Research analyst can keep focusing on the structure of the optimization problem while planners have a powerful tool to plan and visualize the results.