GoogleApi.ContentWarehouse.V1.Model.ResearchScamNeighborSelectionOverride
Table of Contents ▼
Jump to a specific part of the page:
Description
Last used tag = 13
Attributes List
This module has the following attributes (case-insensitive ascending order):
Attributes
-
approxEpsilonDistance
(type:number()
, default:nil
)
- The distance threshold to use for approximate search before exact reordering is performed, if exact reordering is performed. If this is not set and exact reordering is enabled, a reasonable default value will be chosen using a heuristic specified in ScamConfig.ExactReordering.NeighborSelectionOverrideHeuristics. This field is ignored if exact reordering is not enabled. This value must be non-NaN if set. -
approxNumNeighbors
(type:integer()
, default:nil
)
- The number of neighbors to find via approximate search before exact reordering is performed. If this is not set and exact reordering is enabled, a reasonable default value will be chosen using the heuristic specified in ScamConfig.ExactReordering.NeighborSelectionOverrideHeuristics. This field is ignored if exact reordering is not enabled. This value must be > 0 if set. -
epsilonDistance
(type:number()
, default:nil
)
- The maximum distance at which to return a neighbor. If this proto is used, at least one of this and num_neighbors must be set. The default is infinity, effectively returning num_neighbors neighbors regardless of distance. This value must be non-NaN if set. -
numNeighbors
(type:integer()
, default:nil
)
- The maximum number of neighbors to return. If this proto is used, at least one of this and epsilon_distance must be set. The default is the largest value representable as an int32, effectively returning all neighbors within epsilon_distance. This value must be > 0 if set. -
numSingleShardNeighbors
(type:integer()
, default:nil
)
- If populated, this is the maximum number of neighbors that may be returned from a single shard. If not populated, up to num_neighbors neighbors are returned from each shard. Setting this to a smaller value than num_neighbors will increase speed at the expense of accuray by requiring cross-shard merging of fewer neighbors. If this value is set, num_neighbors must also be set and this value must be <= num_neighbors and >= num_neighbors / number of shards. If set, this value must always be > 0. -
perCrowdingAttributeNumNeighbors
(type:integer()
, default:nil
)
- The maximum number of neighbors to return for a single value of the crowding attribute. The crowding attribute is specified per-datapoint in the GenericFeatureVector proto, or the Document proto for sparse logistic models. Crowding is described more thoroughly in research/scam/utils/crowding_top_n.h. NOTES: Crowding is effectivley enabled if this value is less than num_neighbors. -
perCrowdingAttributePreReorderingNumNeighbors
(type:integer()
, default:nil
)
- The maximum number of neighbors to return from the approximate portion of the nearest-neighbor search algorithm, within each shard, for a single value of the crowding attribute, before performing exact reordering. Ignored if exact reordering is disabled for this dataset. -
scoringExtensionMetadata
(type:GoogleApi.ContentWarehouse.V1.Model.ResearchScamScoringExtensionMetadata
, default:nil
)
- The scoring extension metadata overrides for single machine classical scam database. -
treeXHybridCentersToSearch
(type:list(integer())
, default:nil
)
- Note: currently not implemented for all database shard subclasses! Contact ScaM before using. Provides prematched centers_to_search for tree-X hybrid searchers. Ignores any overwrites related to centroid matching since they are already matched.tree_x_hybrid_centers_to_search
should match by size withtree_x_hybrid_distances
. -
treeXHybridDistances
(type:list(number())
, default:nil
)
- -
treeXHybridLeavesSearchedOverride
(type:integer()
, default:nil
)
- Note: currently not implemented for all database shard subclasses! Contact ScaM before using. Overrides max_spill_centers for tree-X hybrid searchers configured with FIXED_NUMBER_OF_CENTERS spilling. The max_spill_centers in the ScaM config is used if this field isn't set. Only valid if: Used on a tree-X hybrid ScaM config with FIXED_NUMBER_OF_CENTERS. Value is greater than zero and at most equal to num_children.
Type
@type t() :: %GoogleApi.ContentWarehouse.V1.Model.ResearchScamNeighborSelectionOverride{
approxEpsilonDistance: number() | nil,
approxNumNeighbors: integer() | nil,
epsilonDistance: number() | nil,
numNeighbors: integer() | nil,
numSingleShardNeighbors: integer() | nil,
perCrowdingAttributeNumNeighbors: integer() | nil,
perCrowdingAttributePreReorderingNumNeighbors: integer() | nil,
scoringExtensionMetadata: GoogleApi.ContentWarehouse.V1.Model.ResearchScamScoringExtensionMetadata.t() | nil,
treeXHybridCentersToSearch: [integer()] | nil,
treeXHybridDistances: [number()] | nil,
treeXHybridLeavesSearchedOverride: integer() | nil
}
approxEpsilonDistance: number() | nil,
approxNumNeighbors: integer() | nil,
epsilonDistance: number() | nil,
numNeighbors: integer() | nil,
numSingleShardNeighbors: integer() | nil,
perCrowdingAttributeNumNeighbors: integer() | nil,
perCrowdingAttributePreReorderingNumNeighbors: integer() | nil,
scoringExtensionMetadata: GoogleApi.ContentWarehouse.V1.Model.ResearchScamScoringExtensionMetadata.t() | nil,
treeXHybridCentersToSearch: [integer()] | nil,
treeXHybridDistances: [number()] | nil,
treeXHybridLeavesSearchedOverride: integer() | nil
}
Function
@spec decode(struct(), keyword()) :: struct()Data sourced from HexDocs : GoogleApi.ContentWarehouse.V1.Model.ResearchScamNeighborSelectionOverride