I have a very basic question but I cannot find an answer (especially for a clustering situation). I am trying to do hierarchical clustering of samples using jaccard distances. One sample contains fewer observations but they are all present in another sample. However, when I calculate the jaccard distance, this is not taken into account and two samples are very far away. How do you take into account this? Is there another way to use somehow make that weight more in the clustering? thanks!


Maybe you used Jaccard similarity rather than Jaccard distance...

The subset property alone shouldn't give a small distance.

Consider a set with 100 items, and a one element subset of it. That is a subset but it's also almost completely dissimilar isn't it?

| cite | improve this answer | |
  • $\begingroup$ I agree partially. let say I have 3 samples: 2 with 100 variables and 1 with 50 variables. Now, between the first two samples, 50% are identical, but all the 50 variables in the third are present in the first. Which one is more similar to the first? The second or the third? I guess for jaccard is the same....and I am not so sure I agree $\endgroup$ – Jordan Jan 12 '19 at 17:46
  • $\begingroup$ No. Jaccard says 1/3 is the same for the first, 1/2 for the second. $\endgroup$ – Has QUIT--Anony-Mousse Jan 12 '19 at 18:37

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.