kumulant

kumulant/com.eignex.kumulant.stat.score/ConfusionMatrixResult

ConfusionMatrixResult

@Serializable

@SerialName(value = "ConfusionMatrixResult")

data class ConfusionMatrixResult(val numClasses: Int, val counts: DoubleArray) : Result(source)

Snapshot of a weighted K-by-K confusion matrix indexed as counts[predicted][truth].

Exposes the family of confusion-matrix metrics as derived getters: per-class precision/recall/F1, macro and micro averages, accuracy, and Matthews correlation coefficient (defined for any K). Empty rows or columns yield 0 for the corresponding rate rather than NaN, which keeps macro averages well-defined while a stream is warming up.

Constructors

ConfusionMatrixResult

constructor(numClasses: Int, counts: DoubleArray)

Properties

val accuracy: Double

Fraction of correctly classified weight; 0 on an empty stream.

val correct: Double

Sum of the diagonal (correctly classified weight).

val counts: DoubleArray

Row-major numClasses * numClasses weights: counts[pred * numClasses + truth].

val macroF1: Double

Macro F1: unweighted mean of per-class F1.

val macroPrecision: Double

Macro precision: unweighted mean of per-class precision.

val macroRecall: Double

Macro recall: unweighted mean of per-class recall.

val mcc: Double

Matthews correlation coefficient generalised to K classes (Gorodkin 2004). Returns 0 when either margin is degenerate.

val numClasses: Int

Number of classes.

val totalWeights: Double

Total weight across all cells.

Functions

fun actualTotal(c: Int): Double

True-class total: weight of all columns where truth = c.

fun count(predicted: Int, truth: Int): Double

counts[pred][truth].

open operator override fun equals(other: Any?): Boolean

fun f1(c: Int): Double

Per-class F1: harmonic mean of precision and recall; 0 when both are 0.

open override fun hashCode(): Int

fun precision(c: Int): Double

Per-class precision: TP / (TP + FP); 0 when no predictions for class c.

fun predictedTotal(c: Int): Double

Predicted-class total: weight of all rows where prediction = c.

fun recall(c: Int): Double

Per-class recall: TP / (TP + FN); 0 when no truth observations for class c.

ConfusionMatrixResult

constructor(numClasses: Int, counts: DoubleArray)(source)

accuracy

val accuracy: Double(source)

Fraction of correctly classified weight; 0 on an empty stream.

actualTotal

fun actualTotal(c: Int): Double(source)

True-class total: weight of all columns where truth = c.

correct

val correct: Double(source)

Sum of the diagonal (correctly classified weight).

count

fun count(predicted: Int, truth: Int): Double(source)

counts[pred][truth].

counts

val counts: DoubleArray(source)

Row-major numClasses * numClasses weights: counts[pred * numClasses + truth].

equals

open operator override fun equals(other: Any?): Boolean(source)

f1

fun f1(c: Int): Double(source)

Per-class F1: harmonic mean of precision and recall; 0 when both are 0.

hashCode

open override fun hashCode(): Int(source)

macroF1

val macroF1: Double(source)

Macro F1: unweighted mean of per-class F1.

macroPrecision

val macroPrecision: Double(source)

Macro precision: unweighted mean of per-class precision.

macroRecall

val macroRecall: Double(source)

Macro recall: unweighted mean of per-class recall.

mcc

val mcc: Double(source)

Matthews correlation coefficient generalised to K classes (Gorodkin 2004). Returns 0 when either margin is degenerate.

numClasses

val numClasses: Int(source)

Number of classes.

precision

fun precision(c: Int): Double(source)

Per-class precision: TP / (TP + FP); 0 when no predictions for class c.

predictedTotal

fun predictedTotal(c: Int): Double(source)

Predicted-class total: weight of all rows where prediction = c.

recall

fun recall(c: Int): Double(source)

Per-class recall: TP / (TP + FN); 0 when no truth observations for class c.

totalWeights

val totalWeights: Double(source)

Total weight across all cells.