How to efficiently get a cluster ID given a sample ID of which sample has been added to a IVF index?

[yzou2]

Currently I have an ID of a vector which has been added to a IVF index and would like to get the cluster assignment. A naive way is firstly defining a mapping from the sample ID to cluster ID given the inverted list and then use the mapping to find the cluster assignment. But is there a build-in function to support fast mapping from sample ID to cluster ID without looking to the inverted list?

[mdouze]

There is no difference between the cluster ID and the inverted list.
You can get the cluster IDs with index.quantizer.search(vector, 1)[1].

If I misunderstood the question, feel free to correct.

[asadoughi]

Re-phrasing the question as: is there an O(1) way of getting the cluster assignment for a given vector id? Invoking search() requires a k-NN computation.

[mdouze]

Right, you need a direct map for that see doc here:
https://github.com/facebookresearch/faiss/wiki/Special-operations-on-indexes#direct-map-for-an-indexivf
and demo here: issue_3555.ipynb