vector search Query with filters

[patelprateek]

Thanks for this amazing work.
Can you please provide some details on current design for query planner when we want to do vector search with some filtering constraints, i observed a pr #1003 which might be relevant , but trying to understand a bit on implementation details here
if the cardinality of query was too low , i assume the previous latency were high because scanning hnsw can be expensive ? why not just do brute force ?
Can you also elaborate a bit on query planner cardinality estimation , how does this really work , how do we estimate cardinality without actually running the filter

[generall]

Hi @patelprateek, thanks for the question. We are not putting those details into the documentation, because details of execution are changing constantly and we wouldn't have a resource to maintain documentation like this. And in my opinion - no documentation is better than outdated documentation.

To answer your question I would refer you to the code, where this logic is defined:

• Search query planner -

  qdrant/lib/segment/src/index/hnsw_index/hnsw.rs

  Line 217 in 516dcd7

  fn search(

• Carnality estimation -

  qdrant/lib/segment/src/index/query_estimator.rs

  Line 90 in 026bd04

  pub fn estimate_filter<F>(estimator: &F, filter: &Filter, total: usize) -> CardinalityEstimation