PerfMeas (version 1.2.1)

PXR: Precision at a given recall level measures

Description

Set of functions to compute the precision at fixed recall levels.

Usage

precision.at.recall.level(scores, labels, rec.level = 0.2)
precision.at.recall.level.over.classes(target, predicted,   g, rec.level = 0.2, root = "00")
precision.at.multiple.recall.level(scores, labels,  rec.levels = seq(from = 0.1, to = 1, by = 0.1))
precision.at.multiple.recall.level.over.classes(target,  predicted, rec.levels = seq(from = 0.1, to = 1, by = 0.1))
precision.at.all.recall.levels(scores, labels)

Arguments

scores

vector of the predicted scores in [0,1]

labels

0/1 vector of the true labels

rec.level

rec.level: the desired recall level (def: 0.2)

target

matrix with the target multilabels; rows correspond to examples, columns to classes

predicted

matrix with the predicted multilabels; rows correspond to examples, columns to classes

g

graph of the classes (object of class graphNEL, package graph). If missing, no per level results are computed.

root

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

rec.levels

a vector with the desired recall levels (def. 0.1 to 1 by 0.1 step)

Value

precision.at.recall.level returns the precision at the requested recallprecision.at.recall.level.over.classes a list with three elements:

- average

the average precision at a given recall level across classes.

- per.level

a named vector with average precision at a given recall level for each level of the hierarchy; names correspond to levels

- per.class

a named vector with precision at a given recall level for each class. Names correspond to classes

precision.at.multiple.recall.level a list with 2 elements:

- precisions

a vector with the precision at different recall levels

- f.score

a vector with the f-score at different recall levels

precision.at.multiple.recall.level.over.classes

- PXR

a matrix with the precisions at different recall levels: rows are classes, columns precisions at different recall levels

- avgPXR

a vector with the the average precisions at different recall levels across classes

precision.at.all.recall.levels a list with 3 elements:

- precision

precision at different thresholds

- recall

recall at different thresholds

- f.score

f.score at different thresholds

Details

precision.at.recall.level computes the precision at a given recall level for a single class.

precision.at.recall.level.over.classes computes precision at a given recall level for a set of classes.

precision.at.multiple.recall.level computes the precision at multiple levels of recall for a single class.

precision.at.multiple.recall.level.over.classes computes the precision at multiple levels of recall for multiple classes.

precision.at.all.recall.levels compute the precision at all recall levels for a single class. It returns a pair of precision and recall values by moving a threshold from the lowest to the highest score: a number of precision and recall values equal to the number of available examples is returned.

See Also

AUC.measures, F.measures

Examples

# preparing pseudo-random predictions 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"];
# precsion at 0.4 recall level for class A
precision.at.recall.level(scores, labels, rec.level=0.4);
# precision at 0.4 recall level for all the 10 classes
precision.at.recall.level.over.classes(Targets, Scores, rec.level=0.4);
# precision at multiple recall levels for class A
levels <- seq(from=0.1, to=1, by=0.1);
precision.at.multiple.recall.level(scores, labels, rec.levels=levels);
# precision at multiple recall levels for all the 10 classes
precision.at.multiple.recall.level.over.classes(Targets, Scores);
# precision, recall and f-score for a single class obtained 
# by moving the threshold across the examples
precision.at.all.recall.levels(scores, labels);