Skip to content

f_new

FNew

Bases: MultiClassMetric

Metric F_new, which represents the percentage of known classes misclassified as novel.

Parameters:

Name Type Description Default
known_classes list of int

List of known labels, the labels the algorithm knows prior to the online phase

 required
cm ConfusionMatrixNovelty

Optional, can specify an already existing confusion matrix instead of creating a new one for the metric

 None

Attributes:

Name Type Description
cm ConfusionMatrixNovelty

Confusion matrix
Source code in streamndr/metrics/f_new.py 
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
class FNew(metrics.base.MultiClassMetric):
    """Metric F_new, which represents the percentage of known classes misclassified as novel.

    Parameters
    ----------
    known_classes : list of int
        List of known labels, the labels the algorithm knows prior to the online phase
    cm : ConfusionMatrixNovelty
        Optional, can specify an already existing confusion matrix instead of creating a new one for the metric

    Attributes
    ----------
    cm : ConfusionMatrixNovelty
        Confusion matrix
    """
    def __init__(self, known_classes, cm: ConfusionMatrixNovelty = None):
        if cm is None:
            cm = ConfusionMatrixNovelty(known_classes)

        super(metrics.base.MultiClassMetric, self).__init__(cm)

    def get(self):
        fp = self.cm.false_positives_novelty() #Number of known class samples wrongly classified as novelties

        try:
            return fp / (self.cm.n_samples - self.cm.nc_samples)
        except ZeroDivisionError:
            return 0.0