Contents   |   Previous

5. Building a bayesian network using Model Construction Wizard

We saw in detail using Bayonet system editor to construct a bayesian network model, in the chapter Building a bayesian network using Bayonet Editor. Bayonet system offers another easy to use tool, Model Construction Wizard which leads to obtain a bayesian network model from experience data. We can start the wizard from the menu, Wizard. There are different menu items available for constructing bayesian network model, based on the behaviour. Presently available commands and methods are,
  1. Model Construction Wizard... - same methods as using the Bayonet editor
  2. Model Construction - Greedy Search... - possible to search for a model using a criteria based on K2 or C4.5
  3. Model Construction - Total Search... - uses greedy serach for preliminary selection of parent nodes and completes constructing model using the Bayonet Tree selection algorithms.
In addition, there is one more wizard for editing the data. That is the Data Preparation Wizard.

Let us look in detail using the wizard. In this tutorial, the command 'Model Construction Wizard...' is explained. The other two can be used similar to this one. Let us use the same database as in the previous chapter. Please read the description about the data and about starting Bayonet system in the previous chapter.

An additional feature in this wizard to the Bayonet editor is that we can use a previously stored model information to build the bayesian network model. Suppose, if you have used the Model Construction - Greedy Search... wizard earlier, you could save the result information in a file and use the file with Model Construction Wizard...

Start Wizard
Create a new project. Then, start the wizard from the menu, Wizard - Model Construction Wizard... .
Click on 'Next' button, in each step to continue untill the last step. If you want to revise something in the previous steps, click on 'Prev' button. Click on 'Cancel' to cancel the wizard.

Step 1
Select the data base type. Let us select File DST for this tutorial.

Step 2
Select the data base. For File DST, select the file containing the experience data.
If you have selected Database DST, specify the database information and click on 'Connect DB'. Then select the table you want to use and fill in the query.

Step 3
Now all the possible node candidates from the database and their properties are displayed. Select the nodes that you want to include in the model being constructed. The node type property, Discrete or Continuous is displayed. If you want to force the system to treat the continuous node as discrete node, uncheck the 'continuous' check box.
Here we can instruct the system to use previously stored information by specifying the file path in the field, 'Use Model Info. from '. The model information file can be saved when using the Greedy Search wizard or if you like, you can create such files yourself. The file should be a text file with lines in the following format.
child node,parent node1,parent node2,parent node3,...
One line represents the information for one tree in the bayesian network model.
Let us select age, job and q1_1_1 for our example. Uncheck on the 'create' column header checkbox to uncheck all nodes.

Step 4
In this step, you can provide your domain knowledge to the wizard, by specifying whether a node has parents and which are the possible parents (click on details column).

Step 5
Determine whether you want to make use of the Bayonet system's Tree selecting algorithms to find the best tree. If you don't want continue to next step and finish the process. Otherwise, check the checkbox for creating all sub models and continue.

Step 6
Now depending on the previous action, you can select an algorithm or simply complete the process by clicking on 'Finish' button.

Save Project
Now the model is displayed on the Bayonet editor. Save this project using 'Save' from the 'File' menu.


Contents   |   Previous