RuleClassifier.R

## Custom implementation of a rule based classificator with rules extracted
## by analyzing one of the optimal Decision Trees

predictByRule <- function(features) {
    em <- rep(FALSE, 7)
    c1  <- features$X87  < -0.58
    c2  <- features$X117 < -1.7157 # Original: -1.7
    c3  <- features$X102 < -0.049
    c4  <- features$X118 < -0.85
    c5  <- features$X120 < -0.81
    c6  <- features$X125 <  0.69
    c7  <- features$X92  <  0.86
    c8  <- features$X89  <  0.86
    c9  <- features$X101 >= -0.014
    c10 <- features$X123 >= -1.4
    c11 <- features$X52  >= -0.61
    
    # Surprise ~ 0% confusion
    em[7] <- c1 && c3
    # Disgust ~ 5,77% confusion
    em[3] <- !c1 && !c2 && c4 && !c8
    # Happiness ~ 5,88% confusion
    em[5] <- !c1 && c2 && c5
    # Sadness ~ 8% confusion
    em[6] <- (c1 && !c3 && c6) || (!c1 && !c2 && !c4 && c7 && c9 && c10 && c11)
    # Fear ~ 9,09% confusion
    em[4] <- (c1 && !c3 && !c6) || (!c1 && !c2 && c4 && c8) || (!c1 && !c2 && !c4 && c7 && !c9)
    # Anger ~ 11,63% confusion 
    em[1] <- (!c1 && !c2 && !c4 && !c7) || (!c1 && !c2 && !c4 && c7 && c9 && c10 && !c11)
    # Contempt ~ 17,65% confusion
    em[2] <- (!c1 && c2 && !c5) || (!c1 && !c2 && !c4 && c7 && c9 && !c10)
    em
}

## Tests performance of the custom rule based classifier, implemented within the predictByRule() function

testRuleClassifier <- function(inputs, trueLabels) {
    hitsNum  <- 0
    falses <- c()
    
    for (inputInd in 1: nrow(inputs)) {
        emotions <- predictByRule(inputs[inputInd, ])
        if (emotions[trueLabels[inputInd]] == TRUE) {
            hitsNum <- hitsNum + 1	
        } else {
            falses <- c(falses, inputInd)
        }
    }
    print(falses)
    hitsNum
}

## [not_finalized] 
## Prediction based on Decision Tree and custom rules. The function
## wraps the tree based predictor, applying manually specified rules when applicable 

predictByTreeRule <- function(DTClassifier, data) {
    classesProbs <- predict(DTClassifier, data)
    maxProb <- max(classesProbs)
    maxInd <- which(classesProbs == maxProb, arr.ind = TRUE)
    
    if (maxInd == 3) {
        for (i in length(classesProbs)) {
            classesProbs[i]
        }
    }
}