rje (version 1.12.1)

conditionMatrix: Find conditional probability table

Description

Given a numeric array or matrix (of probabilities), calculates margins of some dimensions conditional on particular values of others.

Usage

conditionMatrix(
  x,
  variables,
  condition = NULL,
  condition.value = NULL,
  dim = NULL,
  incols = FALSE,
  undef = NaN
)
conditionTable(
  x,
  variables,
  condition = NULL,
  condition.value = NULL,
  undef = NaN,
  order = TRUE
)
conditionTable2(x, variables, condition, undef = NaN)

Value

conditionTable returns an array whose first length(variables) corresponds to the dimensions in variables, and the remainder (if any) to dimensions in condition with a corresponding entry in condition.value of length > 1.

conditionTable2 always returns an array of the same dimensions as x, with the variables in the same order.

Arguments

x

A numeric array.

variables

An integer vector containing the margins of interest from x.

condition

An integer vector containing the dimensions of x to condition on.

condition.value

An integer vector or list of the same length as condition, containing the values to condition with. If NULL, then the full conditional distribution is returned.

dim

Integer vector containing dimensions of variables. Assumed all binary if not specified.

incols

Logical specifying whether not the distributions are stored as the columns in the matrix; assumed to be rows by default.

undef

if conditional probability is undefined, what should the value be given as

order

logical - if TRUE conditioned variables come last, if FALSE variables are in original order.

Functions

• conditionMatrix: Conditioning in matrix of distributions

• conditionTable2: Conditioning whilst preserving all dimensions

Author

Mathias Drton, Robin Evans

Details

conditionTable calculates the marginal distribution over the dimensions in variables for each specified value of the dimensions in condition. Single or multiple values of each dimension in condition may be specified in condition.value; in the case of multiple values, condition.value must be a list.

The sum over the dimensions in variables is normalized to 1 for each value of condition.

conditionTable2 is just a wrapper which returns the conditional distribution as an array of the same dimensions and ordering as the original x. Values are repeated as necessary.

conditionMatrix takes a matrix whose rows (or columns if incols = TRUE) each represent a separate multivariate probability distribution and finds the relevant conditional distribution in each case. These are then returned in the same format. The order of the variables under conditionMatrix is always as in the original distribution, unlike for conditionTable above.

The probabilities are assumed in reverse lexicographic order, as in a flattened R array: i.e. the first value changes fastest: (1,1,1), (2,1,1), (1,2,1), ..., (2,2,2).

condition.table and condition.table2 are identical to conditionTable and conditionTable2.

See Also

marginTable, margin.table, interventionTable

Examples

x = array(1:16, rep(2,4))
x = x/sum(x) # probability distribution on 4 binary variables x1, x2, x3, x4.

# distribution of x2, x3 given x1 = 1 and x4=2.
conditionTable(x, c(2,3), c(1,4), c(1,2))
# x2, x3 given x1 = 1,2 and x4 = 2.
conditionTable(x, c(2,3), c(1,4), list(1:2,2))

# complete conditional of x2, x3 given x1, x4
conditionTable(x, c(2,3), c(1,4))

# conditionTable2 leaves dimensions unchanged
tmp = conditionTable2(x, c(2,3), c(1,4))
aperm(tmp, c(2,3,1,4))

####
set.seed(2314)
# set of 10 2x2x2 probability distributions
x = rdirichlet(10, rep(1,8))

conditionMatrix(x, 3, 1)
conditionMatrix(x, 3, 1, 2)