Machine Learning with Swift

Calculating the confusion matrix

We'll use a straightforward approach here to calculate the confusion matrix; however, this would not work for multiclass classification. Here, p stands for predicted value, and t is for ground truth:

let pairs: [(Int, Int)] = zip(predictions, yVecTest).map{ ($0.0, $0.1) } 
var confusionMatrix = [[0,0], [0,0]] 
for (p, t) in pairs { 
    switch (p, t) { 
    case (0, 0): 
        confusionMatrix[0][0] += 1 
    case (0, _): 
        confusionMatrix[1][0] += 1 
    case (_, 0): 
        confusionMatrix[0][1] += 1 
    case (_, _): 
        confusionMatrix[1][1] += 1 
    } 
} 
     
let totalCount = Double(yVecTest.count) 

Normalize the matrix by total count:

let normalizedConfusionMatrix = confusionMatrix.map{$0.map{Double($0)/totalCount}} 

As we already know, accuracy is a number ...