Efficacy Score for Data Augmentations

A function for calculating the efficacy score, or eff-score, of a data augmentation prior to training and empirical testing of a model.

This function will calculate the eff-score of a given augmentation as described in:

Heise, D. and Bear, H, "A Transparent Method for Visualizing and Scoring the Efficacy of Data Augmentations on Real-World Audio", submitted to IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, in review, 2023.

Inputs:

D - a square dissimilarity matrix, computed via the Visual Assessment of cluster Tendency (VAT) algorithm, of all the samples in the source data

k - an integer specifying the size of neighbourhoods (in pixels) to use for computing diversity (optional; by default, k = 3)

b - an integer specifying the number of class intervals to use for binning the set of diversity values, such as for producing a histogram (optional; by default, b = 20)

w - an integer specifying the width of each class interval (optional, computed automatically if not specified); w should be chosen such that (b*w) is just greater than the maximum diversity

Outputs:

score - the computed eff-score

b_out - the value of b used by the function

w_out - the value of w used by the function

h_fig - a handle to the produced histogram figure