Jaccard calculation about datasketch

[ekzhu]

The first "intersection" you mentioned is not the intersection of set values. It is simply the number of matching hash values between two Minhash. The probability of a single pair of hash values being equal is the same as the Jaccard similarity of two sets. Taking the number of matching pairs of hash values and divide by the total number of hash values, is to compute the estimation of the Jaccard similarity.

See Chapter 3 of Mining of Massive Datasets (http://infolab.stanford.edu/~ullman/mmds/ch3.pdf)

Yes. To compute the exact Jaccard similarity, you take the intersection size and divide by the union size.