Learning | Generate Virtual Database
Updated Feature: Generate Virtual Database
Prior to the 5.4 release, you needed to use the learning algorithm wizards for incorporating prior knowledge into the machine learning process. You had to check the option Keep Structure and define the Number of Structure Equivalent Examples to create a dataset consisting of virtual observations. These virtual observations were generated by drawing from the joint probability distribution encoded by the given Bayesian network, representing your prior knowledge. Then, during the structural learning process, these virtual observations were mixed with the observations from the internal database.
BayesiaLab offers a new item in the Learning menu specifically for creating virtual datasets. This clarifies the approach and extends it to all learning algorithms. The given Bayesian network represents your prior, and you set its weight via Number of Structure Equivalent Examples.
Upon generation of the virtual dataset, the iconappears in the lower right corner of the Graph Window. Hovering over this icon shows you to size of the virtual dataset. You can delete the virtual dataset by left-clicking on the icon. Right-clicking brings up the Bayesian network, i.e. the prior, that generated the data.