(Login) Skip Navigation Links
NetworkExpand Network
 
 
View All Networks View Networks
Help Help Help Documentation
 

Rodent Control Effectiveness Model Tutorial Using CPT Calculator

The lack of or absence of empirical data means that it is often impossible to populate all of the conditional probability tables or CPTs within a Bayesian Network using data. In such cases the only alternative is to use best guesses or expert opinion of the probabilities. A main difficulty associated with this is that CPTs can often be large, making the job of populating them with expert opinion very tedious and prone to inconsistencies.

DBL Interactive contains a built-in CPT Calculator designed to make the job of populating large CPTs using expert opinion simpler. It works by reducing the number of scenarios (rows) in a CPT down to produce a reduced CPT. This reduced CPT contains the minimum number of scenarios required to estimate the influence that each parent node will have on the probabilities in the CPT. You then ask your expert to complete the probabilities for the reduced CPT. Once this is done, you can click on the Calculate CPT button and the CPT Calculator will interpolate probabilities for the full CPT based on the probabilities entered into the reduced CPT. Before the CPT calculator interpolates the full CPT, it checks the logical consistency of the probabilities entered into the reduced CPT. If there are any inconsistencies, it will ask you to adjust the probabilities in the reduced CPT before proceeding.

Lets look at an example. The Rodent Control Effectiveness Model (also listed under Examples) contains a node called Community TBS Effectiveness or F, which has five parent nodes. When this model was built, empirical data for TBS Effectiveness did not exist, so expert opinion was used to populate the CPT for the TBS Effectiveness node. This would ordinarily be a very tedious task because the CPT for the TBS Effectiveness node contains 48 scenarios. Lets look at how the CPT Calculator can make the task of populating this node simpler.

First, open the View Networks page in DBL Interactive. Next, select the  icon beside the Rodent Control Effectiveness Model Using CPT Calculator network. This will present you with a list of all of the nodes contained with the model. Browse to the listing for the Community TBS Effectiveness node and click on the Edit Probabilities link beside it. This will open the CPT for the node. Below the CPT you will see a link called CPT Calculator. Click on this link to open the CPT Calculator (Figure 1).

Figure 1: CPT Calculator Step 1

 
The first step in the CPT Calculator involves ordering the states for the node of interest (the child node) and its parent nodes from most positive state (best state) to most negative state (worst state). In this example, the most positive state for Community TBS Effectiveness is High and the most negative state is Low. There is no middle state so this is set to N/A (not applicable). States for the parent nodes are ordered similarly. For example, to order the states of the parent node Farmer Participation, click on the pencil icon (Edit) beside it and use the drop-down menus to order the states, then click on the save icon (Save). After doing this for all parent nodes (Figure 2), click the View Reduced Probability Table button to open up the reduced probability table (Figure 3).

Figure 2: CPT Calculator Step 1 - ordering states

 
You will notice that the reduced probability table follows a particular format. The first row in the table is the best case scenario. That is, it is the scenario where all parents are in the most positive state. The last row in the table is the worst case scenario. That is, it is the scenario where all parents are in the most negative state. All middle rows in the table represent scenarios where only one parent is not in the most positive state. For example, row two in Figure 3 is where TBS Lure Crops is not in the most positive state (it is Poor) and row three is where TBS Maintenance is not the the most positive state (it is Poor). Click on the pencil icon beside each row in the reduced probability table to edit the probabilities for each scenario.

Figure 3: CPT Calculator Step 2 - reduced probability table

 
The probabilities that you enter into the reduced probability table allow the CPT Calculator to determine the relative influence that a state change in each parent will have on the probabilities in the CPT. It then uses this to interpolate probabilities for all scenarios in the full CPT. Click on the Calculate Probabilities button to run the CPT Calculator and interpolate probabilities for the full CPT of the TBS Effectiveness node. If the probabilities entered into the reduced probability table are logically consistent then the CPT Calculator will finish and allow you to view the calculated probabilities for the node (Figure 4). However, if the probabilities in the reduced probability table are not logically consistent then an error message will appear telling you which probabilities need to be changed.

Figure 4: CPT Calculator Calculated Probabilities

 
For example, in Figure 5, the probability entered in row two for high TBS Effectiveness (1) is greater than the probability entered for the best case scenario in row one (0.9). This is logically inconsistent.

Figure 5: CPT calculator error message caused by logical inconsistency in the reduced probability table
 
DBL Interactive ® v2.0 Copyright 2007 - 2010: School of Integrative Systems, University of Queensland, St Lucia, Australia.