Jaccard similarity for large sets with MinHash