ComputePValue: Compute p-values from information gains and return MDFS

In MDFS: MultiDimensional Feature Selection

Compute p-values from information gains and return MDFS

Description

Compute p-values from information gains and return MDFS

Usage

ComputePValue(
  IG,
  dimensions,
  divisions,
  response.divisions = 1,
  df = NULL,
  contrast.mask = NULL,
  ig.in.bits = TRUE,
  ig.doubled = FALSE,
  one.dim.mode = "exp",
  irr.vars.num = NULL,
  ign.low.ig.vars.num = NULL,
  min.irr.vars.num = NULL,
  max.ign.low.ig.vars.num = NULL,
  search.points = 8,
  level = 0.05
)

Arguments

IG	max conditional information gains
dimensions	number of dimensions
divisions	number of divisions
response.divisions	number of response divisions (i.e. categories-1)
df	vector of degrees of freedom for each variable (optional)
contrast.mask	boolean mask on IG specifying which variables are contrast variables (or NULL if none, otherwise at least 3 variables must be marked)
ig.in.bits	TRUE if input is in binary log (as opposed to natural log)
ig.doubled	TRUE if input is doubled (to follow the chi-squared distribution)
one.dim.mode	'exp' for exponential distribution, 'lin' for linear function of chi-squared or 'raw' for raw chi-squared
irr.vars.num	if not NULL, number of irrelevant variables, specified by the user
ign.low.ig.vars.num	if not NULL, number of ignored low IG variables, specified by the user
min.irr.vars.num	minimum number of irrelevant variables (NULL selects probable optimal number)
max.ign.low.ig.vars.num	maximum number of ignored low IG variables (NULL selects probable optimal number)
search.points	number of points in search procedure for the optimal number of ignored variables
level	acceptable error level of goodness-of-fit one-sample Kolmogorov-Smirnov test (used only for warning)

Value

A data.frame with class set to MDFS. Can be coerced back to data.frame using as.data.frame.

The following columns are present:

• IG – information gains (input copy)

• chi.squared.p.value – chi-squared p-values

• p.value – theoretical p-values

Additionally the following attributes are set:

• run.params – run parameters

• sq.dev – vector of square deviations used to estimate the number of irrelevant variables

• dist.param – distribution parameter

• err.param – squared error of the distribution parameter

• fit.p.value – p-value of fit

Examples

ComputePValue(madelon$IG.2D, dimensions = 2, divisions = 1)

MDFS documentation built on May 31, 2023, 7:31 p.m.