You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We have commands to generate three kinds of graph diagrams - generative graphs, inferential graphs and privacy graphs. At the moment these are fairly basic in structure and could afford to augmented to highlight different properties of the nodes.
I took a look at the typical conventions for displaying generative graphs - they are circles for continuous variables, squares for distinuous variables, double outlines for nodes that follow deterministically from their inputs (regardless of whether they are random variables or not), single outlines for nodes that follow probabilistically. Shaded nodes for observed variables - unshaded for nodes that must be inferred (see Dennis, Lee and Kinnell, 2007, attached description of figure 4 for an example).
Those don't suit us so well. We don't have a way of knowing whether a variable is going to be continuous or discrete until we calculate it. Also, our probabilistic category means that it is a random variable rather than it is derived from a probabilistic calculation.
So let's just say that deterministic nodes have a solid outline. Probabilistic nodes have a dotted outline and nodes that are both have both a dotted and a solid outline.
We have commands to generate three kinds of graph diagrams - generative graphs, inferential graphs and privacy graphs. At the moment these are fairly basic in structure and could afford to augmented to highlight different properties of the nodes.
I took a look at the typical conventions for displaying generative graphs - they are circles for continuous variables, squares for distinuous variables, double outlines for nodes that follow deterministically from their inputs (regardless of whether they are random variables or not), single outlines for nodes that follow probabilistically. Shaded nodes for observed variables - unshaded for nodes that must be inferred (see Dennis, Lee and Kinnell, 2007, attached description of figure 4 for an example).
Those don't suit us so well. We don't have a way of knowing whether a variable is going to be continuous or discrete until we calculate it. Also, our probabilistic category means that it is a random variable rather than it is derived from a probabilistic calculation.
So let's just say that deterministic nodes have a solid outline. Probabilistic nodes have a dotted outline and nodes that are both have both a dotted and a solid outline.
DennisLeeKinnellJML.pdf
The text was updated successfully, but these errors were encountered: