MDFS (version 1.5.5)

ComputePValue: 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
)

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

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)

Examples

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