F-beta Score

This function computes the F-beta score.

Usage

fbeta_score(true, pred, beta = 1, weights = NULL, multi, ...)

Arguments

true

a vector (or a matrix) of observed values. If a matrix is provided, a multi-response is assumed

pred

a vector (or a matrix) of predicted values

beta

a length-one numeric vector, the beta value to consider. Default is beta = 1

weights

observation weights (not implemented yet)

multi

what to do when response has multiple output values

average: errors of multiple outputs are averaged to get a single value for each observation
micro: the score is computed by using the global number of true positives, false negatives and false positives
macro: scores of different classes are averaged by unweighted mean to get a single value. Class imbalance is not taken into account
weighted: scores of different classes are averaged by weighted mean to get a single value. Weights are number of true instances per class. It takes into account label imbalance
raw: returns a vector containing one score for each class
binary: returns the score for the class specified by positive

...

further arguments to multiresponse_classification_metric

Value

A single score for the selected class if multi = "binary", a vector containing one score for each class if multi = "raw", a summary score computed by using the global metrics if multi = "micro" or averaging the results from different classes if multi = "macro". A summary score produced by a weighted average of the results from different class is produced if multi = "weighted"

Details

The F-beta score is a measure of accuracy based on precision and recall. It is defined as

$$F_{\beta} = (1 + \beta^{2}) \frac{precision \times recall}{(\beta^{2} precision) + recall} = (1 + \beta^{2}) \frac{TP}{(1 + \beta^{2})TP + \beta^{2} FN + FP)}$$

The highest possible value of an F-score is 1.0, indicating perfect precision and recall. The lowest possible value is 0.

References

https://en.wikipedia.org/wiki/F-score

Author