Child pages
  • Disjunctive Evidence (5.2)

Contents

Context

Validation Mode | Monitor Tool Bar

New Feature: Disjunctive Inference

Inference in Bayesian network packages is always based on conjunctive inference. We introduced in BayesiaLab 5.2 the possibility to carry out disjunctive inference. 

Upon clicking on  in the Monitor Tool Bar, BayesiaLab switches its inference mode to disjunctive, i.e. the posterior probability distributions and the joint probability are updated according to:

 , for and evidence set made of n observations

instead of the classical conjunctive form:

This inference mode is available for all the tools using exact inference, with Hard and Soft Evidence.

Example

Let's use VisitAsia for illustrating this inference mode.

When the evidence set consist of only one observation, there is obviously no difference between the conjunctive and disjunctive inference.

The observed states are also highlighted with the green bar in the corresponding monitors, but the probability is not 100% as with conjunctive inference but rather the probability of the node given the disjunction:

where is one of the observed node, and denotes its observed state.

Using disjunctive inference, one could then envision a doctor wanting to answer the question:

“What is the probability of getting an abnormal X-ray given the patient has any of these maladies – EITHER Tuberculosis, Cancer OR Bronchitis?” 

We can see in the XRay monitor that it is 15.01%.