AUC measures

Share:

Description

Set of functions to compute the Area Under the ROC Curve (AUC)

Usage

1
2
3

Arguments

pred

numeric vector (scores) of the values of the predicted labels

labels

numeric vector of the true labels (0 negative, 1 positive examples)

target

matrix with the target multilabels: rows correspond to examples and columns to classes. target[i,j] = 1 if example i belongs to class j, target[i,j] = 0 otherwise.

predicted

a numeric matrix with predicted values (scores): rows correspond to examples and columns to classes.

g

a graph of class graphNEL (package graph) of the classes. If g is missing no per.level results are computed

root

the name of the root node (def. "00")

y

a list of lists. The components of the outer list is a list returned from the function AUC.single.over.classes

Details

AUC.single computes the AUC for a single class.

AUC.single.over.classes computes AUC for a set of classes, including their average values across classes and the average values across the levels of the hierarchy (if any); level 1 classes are at distance 1 from the root, level 2 the second level, till to last level correponding to the leaves. Note that if the argument g is missing no per-level values are computed.

compute.mean.AUC.single.over.classes compute means across folds of AUC.single.over.classes. It can be used to automatically computed average values (for each class, level, or average across classes) across folds.

Value

AUC.single returns a numeric value corresponding to the AUC.

AUC.single.over.classes returns a list with three elements:

- average

the average AUC across classes

- per.level

a named vector with average AUC for each level of the hierarchy; names correspond to levels

- per.class

a named vector with AUC for each class; names correspond to classes

compute.mean.AUC.single.over.classes returns a list obtained by averaging the results across folds of the input y. The components are:

- average

the average AUC across classes

- per.level

a named vector with average AUC for each level of the hierarchy; names correspond to levels

- per.class

a named vector with AUC for each class; names correspond to classes

See Also

F.measures, PXR

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
# preparing pseudo.random scores and target-labels for examples: 100 examples
# and 10 classes
Scores <- matrix(runif(1000),nrow=100);
Targets <- matrix(integer(1000),nrow=100);
Targets[Scores>0.5] <- 1;
# adding noise to scores
Scores <- Scores + matrix(rnorm(1000, sd=0.3),nrow=100);
colnames(Scores) <-colnames(Targets) <- LETTERS[1:10];
# getting scores and labels of class "A"
scores <- Scores[,"A"];
labels <- Targets[,"A"];
# AUC for a single class
AUC.single(scores,labels);
# AUC for the 10 classes
AUC.single.over.classes(Targets, Scores);