I want to apply a Naive Bayes/Augmented Naive Bayes algorithm to my data. Although these are very simple algorithms, they give a very good first look at the data. Which is why we often use these. The Target correlation report is a great way to see which nodes are best correlated with the target.Suppose I want apply a p-value cut off (say 0.05) to short list the nodes - is there a way I can do this kind of selection and get that smaller network?Right now I have to manually search for the nodes I want to short list, and remove the rest.
Quote 0 0
Here is a workflow: [list=1:1ma3z59i][*:1ma3z59i]Naïve learning [/*:m:1ma3z59i][*:1ma3z59i]Relationship Analysis (Validation Mode – Analysis – Report). This will return the list of the relations Target-Child, sorted wrt the KL divergence, along with the p-value of the relation[/*:m:1ma3z59i][*:1ma3z59i]Choose the KL divergence value that corresponds to your p-value threshold[/*:m:1ma3z59i][*:1ma3z59i]Use the interactive Arc Force analysis tool (Analysis – Visual – Arc Force) and enter the chosen KL value in the text field (near the slider, at the right of the toolbar)[/*:m:1ma3z59i][*:1ma3z59i]Select all the nodes that satisfy the KL constraint (Ctrl + A)[/*:m:1ma3z59i][*:1ma3z59i]Go back to Modeling mode (F4)[/*:m:1ma3z59i][*:1ma3z59i]Invert the selection (Edit – Invert Selection – Nodes).[/*:m:1ma3z59i][*:1ma3z59i]Exclude the selected nodes or delete them [/*:m:1ma3z59i][/list:1ma3z59i]
Quote 0 0