OneMinusOne{T<:Number} <: TwoClassEncoding{T}

Two class label encoding in which one(T) represents the positive class, and -one(T) the negative class.

OneTwo{T<:Number} <: TwoClassEncoding{T}

Two class label encoding in which one(T) represents the positive class, and 2*one(T) the negative class.

OneVsOne{T} <: TwoClassEncoding{T}

Two class label encoding ...

OneVsRest{T} <: TwoClassEncoding{T}

Two class label encoding ...

OneZero{T<:Number} <: TwoClassEncoding{T}

Two class label encoding in which one(T) represents the positive class, and zero(T) the negative class.

RestVsOne{T} <: TwoClassEncoding{T}

Two class label encoding ...

ConfusionMatrix(targets::AbstractVector, scores::RealVector, thres::RealVector)
ConfusionMatrix(enc::TwoClassEncoding, targets::AbstractVector, scores::RealVector, thres_in::RealVector)

For each threshold from thres computes the binary classification confusion matrix.

ConfusionMatrix(targets::AbstractVector, scores::RealVector, thres::Real)
ConfusionMatrix(enc::TwoClassEncoding, targets::AbstractVector, scores::RealVector, thres::Real)

For the given prediction scores .>= thres of the true labels targets computes the binary confusion matrix.

ConfusionMatrix(targets::AbstractVector, predicts::AbstractVector)
ConfusionMatrix(enc::TwoClassEncoding, targets::AbstractVector, predicts::AbstractVector)

For the given prediction predicts of the true labels targets computes the binary confusion matrix.

precision(args; kwargs...)

Returns precision tp/(tp + fp). Aliases: positive_predictive_value.

accuracy(args; kwargs...)

Returns accuracy `(tp + tn)/(p + n).

auc(x::RealVector, y::RealVector)

Computes the area under curve (x,y) using trapezoidal rule.

balanced_accuracy(args; kwargs...)

Returns balanced accuracy (tpr + tnr)/2.

balanced_error_rate(args; kwargs...)

Returns balanced error rate 1 - balanced_accuracy.

diagnostic_odds_ratio(args; kwargs...)

Returns diagnostic odds ratio tpr*tnr/(fpr*fnr).

error_rate(args; kwargs...)

Returns error rate 1 - accuracy.

f1_score(args; kwargs...)

Returns f1 score 2*precision*recall/(precision + recall).

false_discovery_rate(args; kwargs...)

Returns false discovery rate fp/(fp + tp).

false_negative(args; kwargs...)

Returns # false negative samples.

false_negative_rate(args; kwargs...)

Returns false negative rate fn/p. Aliases: miss_rate, type_II_error.

false_omission_rate(args; kwargs...)

Returns false omission rate fn/(fn + tn).

false_positive(args; kwargs...)

Returns # false positive samples.

false_positive_rate(args; kwargs...)

Returns false positive rate fp/n. Aliases: fall_out, type_I_error.

find_threshold_bins(x::Real, thres::RealVector)

findthresholdbins: x < thres[1] –> 1 thres[i] <= x < thres[i+1] –> i+1 x >= thres[n] –> n+1

fβ_score(args; kwargs...)

Returns fβ score (1 + β^2)*precision*recall/(β^2*precision + recall).

matthews_correlation_coefficient(args; kwargs...)

Returns Matthews correlation coefficient (tp*tn - fp*fn)/sqrt((tp+fp)(tp+fn)(tn+fp)(tn+fn)). Aliases: mcc.

mergesorted(x::Vector{T}, y::Vector{S})

mergesorted(x::Vector{T}, y::T)

negative_likelihood_ratio(args; kwargs...)

Returns negative likelihood ratio fnr/tnr.

negative_predictive_value(args; kwargs...)

Returns negative predictive value tn/(tn + fn).

positive_likelihood_ratio(args; kwargs...)

Returns positive likelihood ratio tpr/fpr.

prcurve(args; kwargs...)

Returns recalls and precisions.

prevalence(args; kwargs...)

Returns prevalence p/(p + n).

quant(args; kwargs...)

Returns quant (fn + tn)/(p + n).

roccurve(args; kwargs...)

Returns false positive rates and true positive rates.

threat_score(args; kwargs...)

Returns threat score tp/(tp + fn + fp). Aliases: critical_success_index.

threshold_at_fnr(targets, scores, fnr)

Returns a decision threshold at a given false negative rate fnr ∈ [0, 1].

threshold_at_fpr(targets, scores, fpr)

Returns a decision threshold at a given false positive rate fpr ∈ [0, 1].

threshold_at_k(scores, k; rev)

Returns a decision threshold at k most anomalous samples if rev == true and a decision threshold at k least anomalous samples otherwise.

threshold_at_tnr(targets, scores, tnr)

Returns a decision threshold at a given true negative rate fpr ∈ [0, 1].

threshold_at_tpr(targets, scores, tpr)

Returns a decision threshold at a given true positive rate tpr ∈ [0, 1].

thresholds(scores; ...)
thresholds(scores, n; reduced, zerorecall)

Returns n decision thresholds which correspond to n evenly spaced quantiles of the given vector of scores. If reduced == true, then the resulting n is min(length(scores) + 1, n). If zerorecall == true, then the largest threshold will be maximum(scores)*(1 + eps()) otherwise maximum(scores).

topquant(args; kwargs...)

Returns topquant 1 - quant.

true_negative(args; kwargs...)

Returns # true negative samples.

true_negative_rate(args; kwargs...)

Returns true negative rate tn/n. Aliases: specificity, selectivity.

true_positive(args; kwargs...)

Returns # true positive samples.

true_positive_rate(args; kwargs...)

Returns true positive rate tp/p. Aliases: sensitivity, recall, hit_rate.