Similarity between bar plots (counts of categories)?

Asked Aug 23 '17 at 15:10

Active Aug 23 '17 at 15:29

Viewed 53 times

What are the best ways to compare (calculate similarity) between two bar plots? For example:

a = {apple: 2, orange: 10, grape: 1}
b = {apple: 1, cucumber: 2, apricot: 3, avocado: 2}
Similarity (a,b) = ?

I want to compare a and b. The things which come to my mind are Jaccard simlarity and Cross Entropy. Are there any better options?

edited Aug 23 '17 at 15:29

asked Aug 23 '17 at 15:10

Michael D

You mean clustered barcharts, don't you? – ttnphns Aug 23 '17 at 15:15
1

See [this](https://stats.stackexchange.com/a/173669/3277) answer and other answers there. – ttnphns Aug 23 '17 at 15:26
1

I meant: you have 2 vectors length 6 with counts (some counts are 0). Consider it as 2x6 frequency cross-table and compute chi-sq distance or phi-sq distance as explained in the answer I linked to. – ttnphns Aug 23 '17 at 15:30
@ttnphns, do you think this is a duplicate? – gung - Reinstate Monica Aug 23 '17 at 15:30
gung, I don't know. That older question is more rich and specific. Maybe a duplicate, maybe not. – ttnphns Aug 23 '17 at 15:32
@ ttnps and @gong, I agree the **chi square distance** is a suitable option. Are there any additional or more sophisticated options? – Michael D Aug 27 '17 at 07:26

0 Answers0