create_prg_curve: Precision-Recall-Gain curve

In Simon-Coetzee/footprintR: StatePaintR: a tool for creating and documenting reproducible chromatin state annotations from StateHub

Description

This function creates the Precision-Recall-Gain curve from the vector of labels and vector of scores where higher score indicates a higher probability to be positive. More information on Precision-Recall-Gain curves and how to cite this work is available at http://www.cs.bris.ac.uk/~flach/PRGcurves/.

Usage

create_prg_curve(labels, pos_scores, neg_scores = -pos_scores)

Arguments

labels	a vector of labels, where 1 marks positives and 0 or -1 marks negatives
pos_scores	vector of scores for the positive class, where a higher score indicates a higher probability to be a positive
neg_scores	vector of scores for the negative class, where a higher score indicates a higher probability to be a negative (by default, equal to -pos_scores)

Details

The PRG-curve is built by considering all possible score thresholds, starting from -Inf and then using all scores that are present in the given data. The results are presented as a data.frame which includes the following columns: pos_score, neg_score, TP, FP, FN, TN, precision_gain, recall_gain, is_crossing and in_unit_square. The resulting points include the points where the PRG curve crosses the y-axis and the positive half of the x-axis. The added points have is_crossing=1 whereas the actual PRG points have is_crossing=0. To help in visualisation and calculation of the area under the curve the value in_unit_square=1 marks that the point is within the unit square [0,1]x[0,1], and otherwise, in_unit_square=0.

Value

A data.frame which lists the points on the PRG curve with the following columns: pos_score, neg_score, TP, FP, FN, TN, precision_gain, recall_gain, is_crossing and in_unit_square. All the points are listed in the order of increasing thresholds on the score to be positive (the ties are broken by decreasing thresholds on the score to be negative).

Simon-Coetzee/footprintR documentation built on May 9, 2019, 1:31 p.m.