com.github.jelmerk.spark.knn.hnsw
HnswSimilarity
Related Doc:
package hnsw
Type of index.
Type of index.
Create the index used to do the nearest neighbor search.
Create the index used to do the nearest neighbor search.
type of the index item identifier
type of the index item vector
type of the index item
type of distance between items
dimensionality of the items stored in the index
maximum number of items the index can hold
the distance function
the distance ordering
invoked for serializing ids when saving the index
invoked for serializing items when saving items
create an index
Creates the model to be returned from fitting the data.
Creates the model to be returned from fitting the data.
type of the index item identifier
type of the index item vector
type of the index item
type of distance between items
identifier
directory containing the persisted indices
number of index partitions
model
Param for the distance function to use.
Param for the distance function to use. One of "bray-curtis", "canberra", "cosine", "correlation", "euclidean", "inner-product", "manhattan" or the fully qualified classname of a distance function Default: "cosine"
Size of the dynamic list for the nearest neighbors (used during the search).
Size of the dynamic list for the nearest neighbors (used during the search). Default: 10
Has the same meaning as ef, but controls the index time / index precision.
Has the same meaning as ef, but controls the index time / index precision. Default: 200
Param that indicates whether to not return the a candidate when it's identifier equals the query identifier Default: false
Param that indicates whether to not return the a candidate when it's identifier equals the query identifier Default: false
Param for the column name for the row identifier.
Param for the column name for the row identifier. Default: "id"
Param to the initial model.
Param to the initial model. All the vectors from the initial model will included in the final output model.
Param for number of neighbors to find (> 0).
Param for number of neighbors to find (> 0). Default: 5
Load an index
Load an index
type of the index item identifier
type of the index item vector
type of the index item
type of distance between items
InputStream to restore the index from
loaded index needs to have space for at least this man additional items
create an index
The number of bi-directional links created for every new element during construction.
The number of bi-directional links created for every new element during construction.
Default: 16
Number of partitions (default: 1)
Number of partitions (default: 1)
Param that specifies the number of index replicas to create when querying the index.
Param that specifies the number of index replicas to create when querying the index. More replicas means you can execute more queries in parallel at the expense of increased resource usage. Default: 0
Param for the output format to produce.
Param for the output format to produce. One of "full", "minimal" Setting this to minimal is more efficient when all you need is the identifier with its neighbors
Default: "full"
Param that specifies the number of threads to use.
Param that specifies the number of threads to use. Default: number of processors available to the Java virtual machine
Param for the partition identifier
Param for the partition identifier
Param for the column name for the query identifier.
Param for the column name for the query identifier.
Param for the column name for the query partitions.
Param for the column name for the query partitions.
Param for the threshold value for inclusion.
Param for the threshold value for inclusion. -1 indicates no threshold Default: -1
identifier
identifier
Nearest neighbor search using the approximative hnsw algorithm.