We have a neardup pipeline at Pinterest, which produces a mapping from every image to a list of up to k near-duplicate images, such as:

near_dups = {

"A": ["B", "I", "K"],

"B": ["A", "D"],

"C": ["E"],

"D": [],

"E": [],

"F": [],

"G": ["K"],

"I": [],

"K": [],

}

Given a mapping such as this one, form neardup clusters: collections of almost identical images.

In the example above, we would form the following groups: (A, B, D, I, G, K), (C, E), and (F).**

Also 3 rapid fire ML questions. Take your time and answer them.