The Belief Function Machine (BFM) is an environment for reasoning with belief functions. This is a new version of the programme (originally written by Phan Giang and Sushila Shenoy in MATLAB and Java), now developed in R - massively using the efficiency of relational databases. The compatibility is guaranteed by using the same file format - AIFF. The current interface is trying to resemble the original look of the tool - and visualize the model using graph with two type of nodes. One type of nodes is used for discrete variables and the other one for the so-called relations. A relation is basically a belief function defined on n-dimensional frame - and connected with respective n variables using edges. Therefore, two nodes of the same type cannot be connected by an edge - at least, id does not make sense. To add nodes and edges to the graph, click on button Edit (disabled now) bellow. After adding a node, you can specify its type, and other properties like name in the menu on the right hand side. To access this menu, click on respective node.
The following two tables serve to definition of respective belief function. The left table lists all focal element - each row corresponds to one focal element - a subset A of the frame . The table has two columns - internal index of the focal element and respectiva mass laid on it. When one clicks on a row - respective focal element is specified in the right hand-side table. Mass laid on selected focal element can be modified easily. Focal elements can be modified, added or removed using auxiliary table called Edited focal Element.
Tip: Choose a variable and press Solve for this variable button. Respective marginal of the model will be computed using fusion algorithm.
Here you can upload your AIFF file with problem definition.
To illustrate the prototype, use one of the following predefined models:
This menu changes based on the type of node you click on.
This is a prototype illustrating different capabilities of an online tool based on R and relational databases. It appears that with a proper coding of a belief function, e.g. Dempster's operator of combination is just a database JOIN, marginalization corresponds to database SELECT and extension of a belief function to a space of a higher dimension can be easily computed using two database JOINs. Because of the well-known efficiency of relational databases, we expect a high computational power of this solution.