buildRegressionTables: Builds conditional probability tables from regressions

In ralmond/CPTtools: Tools for Creating Conditional Probability Tables

Builds conditional probability tables from regressions

Description

Takes a description of a normal graphical model as a series of regressions and a description of the corresponding discrete variables and builds a series of conditional probability tables for the corresponding Bayesian network.

Usage

buildRegressionTables(regs, pvecs, mean = 0, std = 1)

Arguments

regs	A list with names corresponding to the variables in the model giving a series of regression coefficients (see Details).
pvecs	A list with names corresponding to the variables in the model giving a series of probability vectors in order from highest state to lowest state (see Details).
mean	A vector of means for the continuous variables.
std	A vector of standard deviations for the continuous variables.

Details

The regs argument should be a list whose names are the names of the variables. Each element should have the following fields:

b

A vector of slopes for the regression with names equal to the names of the parent variables. Note that if there are no parents, this should be numeric(0).

a

The intercept from the regression.

std

The residual standard deviation from the regression.

The function buildRegressions() creates an appropriate list.

The pvecs should be a list whose names are the names of the variables. Each element should be a named vector of probabilities in order from the Highest to the Lowest state, e.g. c(High=.2,Med=.5,Low=.3).

The values mean and std should either be scalars or vectors of the same length as the number of elements in regs and pvecs. If vectors, they should have names corresponding to the variable names. Note that the order of the elements does not need to be the same in all four arguments, but that the names of all four arguments need to be identical (unless mean or std are given as scalars, in which case they will be appropriately replicated.)

Value

A list of conditional probability tables whose names correspond to regs. Each conditional probability table is expressed as a data frame whose columns correspond to either parent variables or states of the child variable and whose rows correspond to configurations of the parent variable.

Warning

Variables are given from highest to lowest state, for example ‘High’, ‘Medium’, ‘Low’. This is the order expected by StatShop. Note that pvecToCutpoints expects probability vectors in the opposite order.

Author(s)

Russell Almond

References

Almond, R. G. (2010). ‘I can name that Bayesian network in two matrixes.’ International Journal of Approximate Reasoning. 51, 167-178.

Whittaker, J. (1990). Graphical Models in Applied Multivariate Statistics. Wiley.

See Also

buildRegressions

Examples

data(MathGrades)
pl <- buildParentList(structMatrix(MathGrades$var),"Algebra")
rt <- buildRegressions(MathGrades$var,MathGrades$means,pl)
tabs <- buildRegressionTables(rt, MathGrades$pvecs, MathGrades$means,
                              sqrt(diag(MathGrades$var)))

ralmond/CPTtools documentation built on Dec. 27, 2024, 7:15 a.m.