Okapi scoring in sharded clusters

Hi, we are currently migrating from ES and we wanted to know if Atlas Search will allow us to do some custom configuration into the scoring without using the constant scoring.
The cases we want to customize are:

  1. Ignore the term frequency on the scoring. On ES this can be done by customizing the parameters of the algorithm, by setting the algorithm to boolean or using a constant score. Can we do the first solution on Atlas Search?
  2. Do the scoring taking into account the whole collection, not only the current shard. On ES, we can add ?search_type=dfs_query_then_fetch.

Thank you!

Searching on docs for another thing, already discovered the way to do the first case.

For string fields only, it can be set the index config to indexOptions=docs to ignore the term frequency and position (see docs here), which acts like the parameter k on Elasticsearch. With norms=omit we can also ignore the field length, so it’s like the b parameter from Elasticsearch.

Currently, there is no support for calculating the frequencies for the whole collection, so I opened a case for enhancement here.

1 Like

This topic was automatically closed 5 days after the last reply. New replies are no longer allowed.