fit.BN: Fit a Bayesian Network model

In vincenzocoia/copsupp: Vine Copulas made easy

Description

Finds copula models for each pair of response and predictor in some order, so that the pair is conditional on previous predictors.

Usage

fit.BN(
  y,
  xdat,
  ymarg = identity,
  xmargs = identity,
  xord = NULL,
  cops = NULL,
  families = c("bvncop", "bvtcop", "mtcj", "gum", "frk", "joe", "bb1", "bb7", "bb8"),
  ...
)

Arguments

y	Vector of response observations.
xdat	Matrix of predictors. Columns are variables, rows are observations (that correspond with entries of y).
ymarg	Vectorized function of the cdf of the response.
xmargs	List of vectorized functions of the cdfs of the univariate predictors, in the order of the columns of xdat. Or, a single function if they're all the same.
xord	If you already know what order you want to pair the response and the predictors, put that order here as a vector of "column numbers".
cops	If you already know some of the copula models that you'd like to use corresponding to the variables in xord, put them here. Entries with NA will have a copula model chosen.
families	A vector of copula family names to try fitting (will also consider their rotations/reflections). Limited to those families available in VineCopula package, listed in BiCopSelect.
...	Other arguments to pass to VineCopula::BiCopSelect.

Details

This function first determines the order to pair up the response and predictors in the order of highest partial correlation (by using lm). Then the bivariate copula models are chosen and fitted individually using VineCopula::BiCopSelect.

By "pairing response and predictors in some order", I mean pairing (Y,X1), (Y,X2)|X1, (Y,X3)|(X1,X2), etc, though not necessarily in that order.

For the familyset argument, the default is almost all of the families available. It just doesn't include the Tawn copula families.

Value

A list with three entries:

• $xord: The pairing order, as a vector of the column numbers of xdat.

• $cops: A vector of copula family names corresponding to the pairs in the order in $xord.

• $cparstart: A list of copula parameters corresponding to the families in $cops, which should be taken as starting values for an optimization (obtained by bivariate likelihood). If each entry of the list is a vector of length one, then a vector of those values is returned here instead.

Examples

## Get some simulated data:
set.seed(6277)
p <- 5
A0 <- CopulaModel::Dvinearray(p)
copmat0 <- makeuppertri("frk", p-1, p, "")
cparmat0 <- makeuppertri(3, p-1, p)
dat <- fvinesim(100, A0, copmat0, cparmat0)
y <- dat[, 1]
xdat <- dat[, -1]

## Fit the model:
fit.BN(y, xdat)

vincenzocoia/copsupp documentation built on Aug. 23, 2020, 7:37 a.m.