Consider the Bayesian belief network (Bbn) from Adaptation tutorial. Figure 1 shows the current distribution probability for the net.
Figure 1: Current Distribution Probabilities
Now
go back to "edit mode" and set all the conditional distribution
probability to 1 except "Tuberculosis or cancer" node. Switch
back to "run mode" and push "EM-Learning" icon. The EM-Learning
icon is shown in figure 2.
Figure 2: The EM-Learning Icon
When the icon is pushed the EM learning window appears. Note that the field "Convergence threshold" response to the log-likelihood. Push "Select File" button and choose a file from which the conditional distribution probabilities must be computed. When the file is selected the "OK" button appears as shown in figure 3.
Figure 3: The EM-Learning Window
Pushing the "OK" button starts the EM-algorithm. Based on the case set given in the file the EM-algorithm computes the conditional distribution probability for each node. Figure 4 shows the new conditional distribution probability based on asia.dat file.
Figure 4: The New Conditional Distribution Probabilities
![]() |
HUGIN Expert A/S | , 2000 |