ResearchScamNeighborSelectionOverride

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.

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.

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.

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.

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.

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.

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.

The scoring extension metadata overrides for single machine classical scam database.

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 with tree_x_hybrid_distances.

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.