Prior Samples Icon
Although a Bayesian network is usually learned from data without prior knowledge, any available domain knowledge can also be incorporated. You can use a fully specified network (structure and parameters), either a manually constructed or a machine-learned network, to generate virtual particles and mix them with the actual ones described in your dataset. However, this requires you to “weight” your prior, i.e., to define Prior Samples, the number of particles that will be generated. Together with the actual dataset of observations, BayesiaLab can then learn a new network from the cases.
The icon indicates that Prior Samples are in use. By clicking on the icon, you can remove the Prior Samples from the network.
Previously, Prior Samples were also referred to as “virtual samples” that, taken together, create a “virtual database.”