The set of all findings entered into the nodes of a single Bayes net is referred to as a case. A case usually provides some information about a particular object, person, event, thing, etc.
Netica has facilities for working with cases, including the ability to save a case (i.e. all the current positive findings) from a Bayes net to a file. Later those findings can be re-entered into the Bayes net by reading the file.
Case files may consist of many cases (acting as a database, in which each case is a database record). You can use Netica to learn the CPTs of a Bayes net from such a file of cases. Netica can also generate a file of cases which match the probability distribution of a Bayes net, while taking account of findings currently entered. There are a number of other ways to create multi-case files.
Netica can pass through a file of cases, applying Bayes net inference to each case to generate new information about it, and then saving the case with the additional information to a new case file, which is known as processing cases. It can also use a case file to test the performance of a Bayes net, finding error rates, log loss, etc., which is known as testing a net with cases.
Example 1: In a medical example, each case might correspond to a certain patient. When you want to work with a new patient, you save all the information gathered for the first patient into a case file before removing it from the net, perhaps using the patient’s name as a file name. When it comes time to reconsider the first patient - perhaps some lab results have arrived - you just read that person’s case file.
Example 2:
For another example, each case could be a political riding. The
findings would be details about that riding (such as demographics, poll
statistics, etc.), and the Bayes net could be used to predict the percentage
of votes each political party will get in the next election. As
new information about a riding arrived, its case file would be kept updated.