Netica has the ability to revise the CPTs of nodes to account for the currently entered case, as well as some other learning abilities.
Preparation: To learn from a single case, you must first have a net constructed, including all nodes, states and links. Nodes in the net may already have their CPTs, which you entered manually or previously learned, and which you now want to improve using learning. Or there might not be any CPTs, and you want to learn them from scratch.
Doing: If the case is not already in the Bayes net, you enter it into the net as findings. Only positive findings will be used; negative and likelihood findings will be ignored. Then you select the nodes whose probabilities you would like to have revised to account for the case, if possible. Usually you would like all possible nodes to have their probabilities revised, so you would select all the nodes (or don’t select any nodes, which is equivalent). Then choose Cases → Learn → Incorporate Case, or click the toolbar button which has an arrow pointing from the case symbol to the relation symbol: .
Degree: You will then be queried for a “degree”, which is normally 1. By making it 2, you can achieve the same effect as learning the same case twice, and equivalently for other numbers. By making it -1, you can exactly unlearn a case that was earlier learned with degree = 1, and so on for other negative numbers. Don’t try to unlearn cases that were never learned, or to unlearn them with greater degree than they were learned.
What Happens: Selected nodes
for which the case provides sufficient data (i.e. findings for it and
its parents)
will have their probabilities revised a small amount to account for the
case, and their appropriate experience levels increased slightly, according
to Netica’s learning algorithm.
Home > Learning From Cases > Single Case Learning