CODEWITHSUNDEEP
elasticsearch
HoangND
HoangND

Reputation: 408

Elasticsearch l1norm function score

I'm trying to find out whether l1norm function calculates score the way I think it is.

So I have this index

PUT my-index-000001
{
  "mappings": {
    "properties": {
      "my_dense_vector": {
        "type": "dense_vector",
        "dims": 3
      }
    }
  }
}

PUT my-index-000001/_doc/1
{
  "my_dense_vector": [3, 4, 5]
}

PUT my-index-000001/_doc/2
{
  "my_dense_vector": [1, 2, 3]
}

POST my-index-000001/_refresh

When I execute this query:

GET my-index-000001/_search
{
  "query": {
    "script_score": {
      "query" : {
        "match_all" : {}
      },
      "script": {
        "source": "l1norm(params.queryVector, 'my_dense_vector')", 
        "params": {
          "queryVector": [3, 3, 3]
        }
      }
    }
  }
}

It seems like the scores are calculated as sum of absolute deltas, or this formula

ABS(X1-Y1) + ABS(X2-Y2) + ABS(X3-Y3)

I just want confirm if this is how the scores are calculated for l1norm function

Upvotes: 0

Views: 54

Answers (1)

ilvar
ilvar

Reputation: 5841

Yes, this is how L1 distance is defined

Upvotes: 2

Related Questions