Query Parameters¶

Query parameters control search-time behavior for vector queries. Each index type has its own query parameter class.

Inheritance Hierarchy¶

QueryParams (abstract)
  ├── HnswQueryParams
  ├── IVFQueryParams
  └── FlatQueryParams

Shared by all query parameter types:

Method	Returns	Description
`type`	`IndexType`	Index type this applies to
`radius`	Float	Max distance threshold
`radius=`	—	Set max distance threshold
`linear?`	Boolean	Force linear (brute-force) scan
`linear=`	—	Set linear scan mode
`using_refiner?`	Boolean	Re-rank with exact distances
`using_refiner=`	—	Set refiner mode

qp = Zvec::HnswQueryParams.new(ef: 300, radius: 0.0, is_linear: false, is_using_refiner: false)

Parameter	Type	Default	Description
`ef`	Integer	`300`	Search width at query time
`radius`	Float	`0.0`	Max distance (0 = unlimited)
`is_linear`	Boolean	`false`	Bypass HNSW, use brute force
`is_using_refiner`	Boolean	`false`	Re-rank candidates exactly

Instance methods:

Method	Returns	Description
`ef`	Integer	Search width
`ef=`	—	Set search width

qp = Zvec::IVFQueryParams.new(nprobe: 10, is_using_refiner: false, scale_factor: 10.0)

Parameter	Type	Default	Description
`nprobe`	Integer	`10`	Number of clusters to search
`is_using_refiner`	Boolean	`false`	Re-rank candidates exactly
`scale_factor`	Float	`10.0`	Internal scaling parameter

Instance methods:

qp = Zvec::FlatQueryParams.new(is_using_refiner: false, scale_factor: 10.0)

Parameter	Type	Default	Description
`is_using_refiner`	Boolean	`false`	Re-rank candidates
`scale_factor`	Float	`10.0`	Internal scaling parameter

Instance methods:

Method	Returns	Description
`scale_factor`	Float	Scale factor
`scale_factor=`	—	Set scale factor

Pass query params as the query_params keyword argument:

qp = Zvec::HnswQueryParams.new(ef: 500)
results = col.query_vector("embedding", query_vec, top_k: 10, query_params: qp)