Seminar

F. J. Díez. Elvira: A software package for probabilistic DAGs. Harvard School of Public Health. Boston, October 16, 2002.

16 slides (8 pages). PDF(439 KB).

Abstract

Elvira is a computer program developed as a join project of several Spanish universities for the construction and evaluation of probabilistic directed acyclic graphs (DAGs), such as Bayesian networks and influence diagrams. It has a user-friendly graphical interface, with advanced explanation capabilities (especially for causal models), treatment of canonical interactions (noisy OR, MAX, AND, etc.), and localization facilities (it can be easily "translated" into different languages). Elvira implements some of the standard algorithms for the automatic construction and evaluation of probabilistic DAGs, as well as new algorithms developed in the Elvira project. The source program, made up of around 115,000 lines of Java code, is publicly available on Internet. The main weakness of Elvira is that it is still under development. This presentation will also discuss some medical causal Bayesian networks built with Elvira, such as ProstaNet (urology), NasoNet (oncology), and Hepar (hepatology). He will also discuss the use of Elvira as a pedagogical tool in epidemiology for intuitively understanding the properties of causal probabilistic models, such as d-separation and the activation of backdoor paths.

Related publications