Get Node Distributions from Model Object

[jdutch27]

Within setNode(), we could have a new parameter called modelObject.

---

Error Checking:

Logic/error checking to ensure class(modelObject) %in% c("lm","glm","multinom",<any others?>)

Logic/error checking to ensure all variables showing up in model equation actually are parents of the node in the graph. Something like:

length(setdiff(setdiff(names(modelObject$model), node), parents("d.dimer", net$dag))) == 0

But need to be careful that modelObject$model exists for all types of allowable model classes.

A good test would be to try to do a spline model (e.g., y ~ bs(x1, df=5))

---

Behavior (pseudocode):

modelType <- class(modelObject)

if(modelType)=="lm" {
nodeType <- "dnorm";
nodeFormula <- [ extract fitted regression equation from modelObject ]
} else if(modelType == "glm") {
nodeType <- [logic to see which family was used for the glm]
[behavior slightly varies depending on nodeType - may or may not need to specify error distribution]
} else {
nodeType <- "dcat";
[ set multinomial logistic regression equations using the multinom object ]
}

[nutterb]

Implemented on 20 Feb 2015