RASON® – Restful Analytic Solver® Object Notation – is a modeling language embedded in JSON and a REST API that makes it easy to create, test and deploy decision services powered by analytic models in web and mobile applications – using business rules and DMN decision tables, optimization, simulation, forecasting and machine learning. It's supported by Excel Solver developer Frontline Systems.
You can use RASON as a cloud service – register here for a free trial account – or in SDK form on your desktop or server – register here to download a free trial.
If you have used another modeling language for optimization or simulation, you’ll find the RASON language simple but powerful and expressive – and that integrating RASON models into a larger application, especially a web or mobile app, is much easier than with other modeling systems.
If you have used Excel for optimization, simulation or decision tables, you’ll find that it’s easy to translate Excel models into RASON models, that your knowledge of Excel formulas and functions is immediately usable, but that RASON models can be more flexibly “bound” to data from a variety of sources.
• RASON's decision tables follow the DMN 1.2 standard, with S-FEEL syntax for rules, and inputs from any of RASON's predictive and prescriptive analytics methods.
• RASON's mathematical optimization methods include linear programming and mixed-integer programming, convex quadratic programming and second-order cone programming, smooth nonlinear and global optimization, genetic algorithms and tabu search -- from small to very large (LP/MIP models with millions of variables).
• RASON's Monte Carlo simulation methods include stratified sampling, rank-order correlation and copulas, 100+ distributions and statistics, simulation optimization, robust optimization, and stochastic programming methods. With built-in stochastic decomposition methods, solve problems scaled-up to the state of the art.
• RASON's forecasting and machine learning methods include text mining, partitioning, feature selection, principal components, clustering, ARIMA and exponential smoothing, linear and logistic regression, k-nearest neighbors, classification and regression trees, multi-layer neural networks, and ensembles of most algorithms.