Frequently Asked Questions


What are Bayesian networks?

Bayesian networks, also called Bayesian belief networks or belief networks, are acyclic directed graphs in which vertices (nodes) represent random variables and edges (arcs) represent directed influences between the variables. Bayesian networks are capable of representing causal relationships between variables and are, somewhat imprecisely, also called causal networks. For more information, please read our short introduction to Bayesian networks.


What are GeNIe and SMILE?

SMILE (Structural Modeling, Inference, and Learning Engine) is a fully portable library of C++ classes implementing graphical decision-theoretic methods, such as Bayesian networks and influence diagrams, directly amenable to inclusion in intelligent systems. Its graphical user interface, GeNIe is a versatile and user-friendly development environment for graphical decision-theoretic models. Both modules, developed originally at the Decision Systems LaboratoryUniversity of Pittsburgh, were made available to the community in July 1998 and its user base reached thousands of users worldwide only a decade later.  The software has become world-famous for its quality, friendliness of its user interface, reliability, functionality, and speed.


What are the advantages of GeNIe and SMILE?

There are several companies offering solutions based on Bayesian methods.  Some of them date back to the very early days of work on probabilistic graphical models.  GeNIe and SMILE have become famous among the members of the academic, research, and industrial communities for their unparalleled user interface, functionality, versatility, reliability, and speed.  Written in standard C++, it is fully portable and runs on a variety of platforms, including dedicated on-board hardware.  Proximity to the best research centers at the University of Pittsburgh and Carnegie Mellon University has resulted in the software that is up-to-date on all successful theoretical developments, being at the same time reliable and very fast.


If a decision-analytic model is subjective, how can I test its correctness or accuracy?

There are many ways in which you could address the problem of model correctness and accuracy. If you have access to test cases for which the optimal decisions are known, we suggest that you test your model on these cases. This will give you an excellent opportunity to gain confidence in your model or to revise and improve it. You can confront your domain experts with the results generated by the model and incorporate their opinion in your model. Finally, if you have data sets available, you can test your model’s accuracy through cross-validation, a standard functionality of GeNIe and SMILE.


I have a rich data set of cases. Can I learn a Bayesian network model or its parameters from these data?

Learning is a fundamental capability of a modern Bayesian networks library. GeNIe and SMILE’s learning module allows users to learn both the structure and the parameters from data sets. This is especially useful for those users who have rich data sets, such as insurance companies, banks, industrial sites. Models learned from data can match the problem very accurately without the need for subjective probability estimates. Another application of the learning module is causal discovery, i.e., discovery of possible causal relationships among the measured variables. Knowledge of causal relationships and their strengths allows for predicting the effects of manipulation, i.e., introduction of policies that change the values of some of the modeled variables.


What is in the name BayesFusion?

BayesFusion refers to our desire to combine the highly successful Bayesian framework with existing methods for modeling, such as simultaneous structural equation models, and data analysis, such as a variety of statistical and data mining tools into a comprehensive, hybrid framework that solves practical problems. Such framework is based on the sound foundations of Bayesian probability theory but does not replace time-tested engineering and economic methods, being at the heart of most solutions used at present. The result of this approach is a highly versatile, efficient, and precise framework that gives solid foundation to the best possible solutions of hard problems.