Block: Define parameter groups

View source: R/NeighborhoodEstimation.R

BlockR Documentation

Define parameter groups

Description

Define parameter groups and relative penalization. If this.model is specified, then, neither blist nor nlist should be provided. If this.model is not specified then both of the preceeding must be provided. When this.model is provided, it is assumed to have been created by with makeModel. Column indices indicated in the attribute 'fixedCols' are unpenalized intercepts, then followed by pairs of interaction terms. The grouping corresponds to the parameter set for a node. Otherwise, blist, mlist and nlist should be set. See details.

Usage

Block(
  this.model,
  blist,
  mlist,
  nlist,
  group = "components",
  lambda,
  penalty.scale = NULL
)

Arguments

this.model

(optional) a matrix created by makeModel

blist

a list of parameter indices, one per block. By default the first block is assumed to be unpenalized.

mlist

a list of parameter indices, one per column of the model.matrix. If omitted, assumed to equal to the identity.

nlist

a named list of block indices, one per node

group

character: one of components, or none.

lambda

currently ignored

penalty.scale

optional list containing elements 'scale' and 'group'. ‘group' should be one of ’block' or 'none'. 'scale' should be numeric of length 'blist' or the sum of the 'blist' lengths.

Details

There are four components that all need to be mapped between each other. In increasing abstraction, with variable prefixes in parenthesis: 1. (p)arameter vector. The map is given in these terms. 2. (mm) model matrix–columns from the covariate matrix. 3. (b)locks – penalty groups 4. (n)odes. – the graph-theoretic structure. 5. lambda – penalties as a function of blocks. All of these components are provided in the 'map'

Value

a list containing a data.table 'map' giving the mapping between parameters, groups and penalty scales and some other components


amcdavid/HurdleNormal documentation built on May 14, 2022, 11:12 p.m.