I have discrete (binned) histograms that I want to compare in order to find a similarity in the range 0 to 1 (or 0% to 100%). I will do a hypothesis test with null hypothesis and things like that. So, for 9 histograms I will use all the combinations (35-36 total) and use pair-wise error. I have tried Kolmogorov-Smirnov but it is for continuous histograms. I also tried Chi-squared but it gives me raw numbers. In one comparison it is 200, in another 1000, and the latter does not help. I even used EMD but the same as Chi-squared. Any other idea??
Asked
Active
Viewed 21 times
1
-
1Consider also asking in Stats StackExchange – Benjamin Wang Aug 03 '22 at 08:35
-
1Use. the Kullback-Leibler divergence: https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence – David G. Stork Aug 03 '22 at 09:01
-
@ David G. Stork: Thanks a lot! But, it does not support p-value. – just_learning Aug 03 '22 at 10:28