sevt_fit: Fit a staged event tree

In stagedtrees: Staged Event Trees

Fit a staged event tree

Description

Estimate transition probabilities in a staged event tree from data. Probabilities are estimated with the relative frequencies plus, eventually, an additive (Laplace) smoothing.

Usage

sevt_fit(
  object,
  data = NULL,
  lambda = NULL,
  scope = NULL,
  compute_logLik = TRUE
)

Arguments

object	an object of class sevt.
data	data.frame or contingency table with observations of the variables in object.
lambda	smoothing parameter or pseudocount. Default (NULL) to lambda value stored in object. If no lambda value is stored nor provided, 0 will be used with a warning.
scope	which variable should be fitted. Default (NULL) to all variables in the model. A partial re-fit is possible only for model which are already fitted and in that case the provided lambda will be ignored if different from object$lambda.
compute_logLik	logical value. If TRUE the log-likelihood of the model is computed and stored.

Details

The data in form of contingency tables and the log-likelihood of the model is (eventually) stored in the returned staged event tree. Partial re-fit of a model can be performed with the scope argument. Partial re-fit can only be done over a fully fitted model, e.g. when changing the stages structure of one of the variables. In case of a partial re-fit, the data and lambda arguments will be ignored and the data and lambda value stored in the sevt object will be used (a warning is issued if such arguments are supplied).

Value

A fitted staged event tree, that is an object of class sevt with ctables and prob components. Additionally the chosen lambda is stored in the returned object and eventually the log-likelihood of the model is saved in the ll field.

Examples

#########
model <- sevt(list(
  X = c("good", "bad"),
  Y = c("high", "low")
))
D <- data.frame(
  X = c("good", "good", "bad"),
  Y = c("high", "low", "low")
)
model.fit <- sevt_fit(model, data = D, lambda = 1)

stagedtrees documentation built on May 29, 2024, 12:33 p.m.