Previous Up Next

module Biocaml_roc: sig .. end

Performance measurement for binary classification.

This module provides functions to compute various performance measurement for binary classification. Typically, binary classifiers output both a label and a score indicating a confidence level. A ROC curve represents the variation of sensitivity and specificity of the classifier as a function of a score threshold.

type confusion_matrix = private {

val make  : pos:float Stream.t ->
 neg:float Stream.t -> (float * confusion_matrix) Stream.t

Given an enum pos (resp. neg) of scores from positive (resp. negative) instances, make ~pos ~neg builds an enum of confusion matrices by setting an decreasing acceptance threshold. The result has at least one first value, which is the confusion matrix for an infinity threshold. The subsequent thresholds are the values in pos and neg.

val sensitivity  : confusion_matrix -> float

val false_positive_rate  : confusion_matrix -> float

val accuracy  : confusion_matrix -> float

val specificity  : confusion_matrix -> float

val positive_predictive_value  : confusion_matrix -> float

val negative_predictive_value  : confusion_matrix -> float

val false_discovery_rate  : confusion_matrix -> float

val f1_score  : confusion_matrix -> float

val auc  : (float * float) Stream.t -> float

auc e computes the area above the X-axis and under the piecewise linear curve passing through the points in e. Assumes that the points come with increasing x-coordinate.