RMixtCompUtilities (version 4.1.6)

computeDiscrimPowerVar: Discriminative power

Description

Compute the discriminative power of each variable or class

Usage

computeDiscrimPowerVar(outMixtComp, class = NULL)
computeDiscrimPowerClass(outMixtComp)

Value

the discriminative power

Arguments

outMixtComp

object of class MixtCompLearn or MixtComp obtained using mixtCompLearn or mixtCompPredict functions from RMixtComp package or rmcMultiRun from RMixtCompIO package.

class

NULL or a number of classes. If NULL, return the discriminative power of variables globally otherwise return the discriminative power of variables in the given class

Author

Matthieu Marbac

Details

The discriminative power of variable j is defined by 1 - C(j) $$C(j)= - \sum_{k=1}^K sum_{i=1}^n P(Z_i=k|x_{ij}) \log(P(Z_i=k|x_{ij})) / (n*\log(K))$$

A high value (close to one) means that the variable is highly discriminating. A low value (close to zero) means that the variable is poorly discriminating.

The discriminative power of variable j in class k is defined by 1 - C(j) $$C(j)= - sum_{i=1}^n (P(Z_i!=k|x_{ij}) \log(P(Z_i!=k|x_{ij})) + P(Z_i=k|x_{ij}) \log(P(Z_i=k|x_{ij}))) / (n*\log(2))$$

The discriminative power of class k is defined by 1 - D(k) $$D(k) = -\sum_{i=1}^n P(Z_i=k|x_i) \log(P(Z_i=k|x_i)) / (n*\exp(-1))$$

See Also

plotDiscrimClass plotDiscrimVar

Examples

if (requireNamespace("RMixtCompIO", quietly = TRUE)) {
  dataLearn <- list(
    var1 = as.character(c(rnorm(50, -2, 0.8), rnorm(50, 2, 0.8))),
    var2 = as.character(c(rnorm(50, 2), rpois(50, 8)))
  )

  model <- list(
    var1 = list(type = "Gaussian", paramStr = ""),
    var2 = list(type = "Poisson", paramStr = "")
  )

  algo <- list(
    nClass = 2,
    nInd = 100,
    nbBurnInIter = 100,
    nbIter = 100,
    nbGibbsBurnInIter = 100,
    nbGibbsIter = 100,
    nInitPerClass = 3,
    nSemTry = 20,
    confidenceLevel = 0.95,
    ratioStableCriterion = 0.95,
    nStableCriterion = 10,
    mode = "learn"
  )

  resLearn <-RMixtCompIO::rmcMultiRun(algo, dataLearn, model, nRun = 3)


  discVar <- computeDiscrimPowerVar(resLearn)
  discVarInClass1 <- computeDiscrimPowerVar(resLearn, class = 1)
  discClass <- computeDiscrimPowerClass(resLearn)

  # graphic representation of discriminant variables
  plotDiscrimVar(resLearn)
  # graphic representation of discriminant classes
  plotDiscrimClass(resLearn)
}