sort score labels putting high scores first

in-place sort of score labels; putting high scores first

cumulated actives curve given an already sorted list of score labels

ROC curve (list of (FPR,TPR) values) corresponding to those score labels

same as roc_curve but for an already sorted array of score-labels

Precision Recall curve (list of (recall,precision) pairs) corresponding to given score labels

compute Area Under the ROC curve given an already sorted list of score labels

ROC AUC: Area Under the ROC curve given an unsorted list of score labels

PR AUC: Area Under the Precision-Recall curve given an unsorted list of score labels

Area Under the ROC curve given an unsorted array of score labels; WARNING: array will be modified (sorted)

Area Under the ROC curve given an already sorted array of score labels

(early) enrichment factor at given threshold (database percentage) given an unsorted list of score labels

(early) enrichment factor at given threshold (database percentage) given an already sorted array of score labels

initial_enhancement a score_labels will compute S = sum_over_i (exp (-rank(active_i) / a)) given an unsorted list of score labels. Robust Initial Enhancement (RIE) = S/<S> where <S> is the average S for randomly ordered score labels. RIE = 1.0 <=> random performance. Cf. DOI:10.1021/ci0100144 for details.

same as initial_enhancement but does not reorder the list of score_labels

power metric at given threshold given an unsorted list of score labels

power metric at given threshold for an already decr. sorted list of score labels

bedroc_auc at given alpha. Default alpha = 20.0.

bedroc_auc at given alpha for an array of score-labels. Default alpha = 20.0. WARNING: the array will be modified (sorted by decrasing scores) if sorted = false which is the default.

equivalent to bedroc_auc_a ~alpha ~sorted:true arr.

Matthews' Correlation Coefficient (MCC) use: mcc classif_threshold score_labels. scores >= threshold are predicted as targets; scores < threshold are predicted as non targets.

a, b = platt_scaling ~debug score_labels Fit a logistic curve (1 / (1 + exp (ax + b))) to score_labels and return its (a, b) parameters. Gnuplot it used underneath to do the fitting. Biblio: Platt, J. (1999). Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in large margin classifiers, 10(3), 61-74.

platt_probability a b score transform score into a probability, given logistic function parameters a and b obtained from a prior call to platt_scaling.