Nirvana Distance
This code computes the nirvana distance, or distance from "ideal" for a data augmentation.
This function will calculate the nirvana distance -- that is, the distance from "ideal" -- for a data augmentation.
Inputs:
C - an integer matrix of the confusion data for the augmentation under evaluation, with ground truth labels in rows and predicted labels in columns
F - a square matrix representing the distances between target classes in the original (non-augmented) data feature space, with order of classes as in C
Outputs:
ND - the computed nirvana distance
dc - a vector of values representing the distance component for each target class in the data set
引用格式
D. Heise and H. Bear, "Evaluating the Potential and Realized Impact of Data Augmentations", submitted to 2023 IEEE Symposium Series on Computational Intelligence, in review.
