synthval.metrics

Module for computing various similarity metrics between two sets of samples originating from multivariate distributions.

This module defines abstract and concrete classes for computing similarity metrics between samples from two distributions. The available metrics include Kullback-Leibler divergence, Wasserstein distance, Energy distance, Mean Mahalanobis distance, Frechet Distance, Inception Score, Kernel Distances and others. Given the scope of the package, we identify the first distribution as real_dist and the second as synth_dist indicating the distribution of the real data and the distribution of the synthetic data respectively. It should be noted that the methods of the package can be generally applied to any kind of multivariate distributions.

Classes

SimilarityMetric(abc.ABC)

Abstract base class for defining similarity metrics.

KLDivergenceEstimation(SimilarityMetric)

Concrete implementation of Kullback-Leibler divergence estimation.

WassersteinDistance(SimilarityMetric)

Concrete implementation of the Wasserstein distance.

EnergyDistance(SimilarityMetric)

Concrete implementation of the Energy distance.

MeanMahalanobisDistance(SimilarityMetric)

Concrete implementation of the mean Mahalanobis distance.

FCNNAccuracyMetric(SimilarityMetric)

Concrete implementation of an accuracy metric based on fully-connected neural networks.

InceptionScore()

Evaluation metrics for generated images.

FrechetDistance(SimilarityMetric)

Concrete implementation of the Frechet distance.

KernelDistance(SimilarityMetric)

Concrete implementation of the Kernel distance.

PRScore(SimilarityMetric)

Concrete implementation of the Precision and Recall Scores.

Classes

SimilarityMetric	Abstract base class representing a generic similarity metric between two sets of samples originating from
KLDivergenceEstimation	Similarity Metric computing an estimation of the Kullback-Leibler divergence based on the methodology proposed in
WassersteinDistance	Similarity Metric computing the Wasserstein Distance.
EnergyDistance	Similarity Metric computing the Energy Distance.
MeanMahalanobisDistance	Similarity Metric computing the mean of the estimated Mahalanobis Distances between all the samples
FCNNAccuracyMetric	Similarity Metric computing the Accuracy of a fully-connected neural networks trained to distinguish between the
InceptionScore	Class for computing the Inception Score over a set of probabilities.
FrechetDistance	Similarity Metric that computes the Frechet distance (also known as Fréchet Inception Distance) between two
KernelDistance	Similarity Metric that computes the Kernel distance between two distributions using provided samples.
PRScores	A Similarity Metric class that computes the Precision and Recall scores between two distributions