Is it possible to execute an approximate nearest neighbour search by using the document id as a parameter?

[olsi-qose]

In the docs it is specified that the knn query can be used to search nearest neighbours.

GET my-knn-index-1/_search
{
  "size": 2,
  "query": {
    "knn": {
      "my_vector2": {
        "vector": [2, 3, 5, 6],
        "k": 2
      }
    }
  }
}

The dimensionality of the data is approximately 1300 embeddings. I was wondering if it is possible to query knn results by passing the id of the ES document instead of the whole vector?

[aetter]

Hi @olsi-qose not as far as I know -- I believe you have to get the document you want first and then pass over the appropriate field as your query -- but I've tagged a couple people who are more knowledgeable than me. @jmazanec15 @vamshin.

[olsi-qose]

Thank you @aetter!

[jmazanec15]

Hi @olsi-qose, at the moment, we do not support this. But it is an interesting use case. I will create an issue in the k-NN repo for investigation.

[aetter]

Closing this out in favor of opendistro-for-elasticsearch/k-NN#341.