I have multiple 20x20 matrices that represent histogram distributions. I want to compare these matrices with one another, and I'd prefer a metric that would return a scalar metric that correlates to the degree of similarity between two histograms other than a euclidean L"n" distance. Similarity can be proportional or inversely to the magnitude of the scalar, as long as it's consistent. Any ideas? I know similarity is very ambiguous, but I would like somewhere to start.


By "20 X 20 matrices" I assume you mean that you have bi-variate distributions (rows and columns are bin ranges for each of two variables, while cell values are counts for each bin). I also assume that your bin sizes are identical for all matrices.

What you are looking for is a "statistical distance": "A distance between populations can be interpreted as measuring the distance between two probability distributions and hence they are essentially measures of distances between probability measures." https://en.wikipedia.org/wiki/Statistical_distance

The first answer to this question should help you: How to measure similarity of bivariate probability distributions?

Here's another question and answers that has related information: https://stackoverflow.com/questions/10002261/kullback-leibler-divergence-as-histogram-distance-function


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.