Skip to contents

Helper function to generate kNN defaults

params_knn_defaults.Rd

This function generates various sensible default parameters for all of the different approximate nearest neighbours that are available within this package.

Usage

params_knn_defaults()

Value

A list with default parameters for kNN searches. Following parameters:

  • k - Number of neighbours. Defaults to 15L.

  • knn_method - Which of method to use for the approximate nearest neighbour search. Defaults to "hnsw". The implementations are: c("hnsw", "annoy", "nndescent", "ivf", "exhaustive").

  • ann_dist - Which distance metric to use for the approximate nearest neighbour search. Defaults to "cosine". The implementations are c("cosine", "euclidean").

  • n_trees - Annoy param: number of trees to generate for Annoy. Defaults to 50L.

  • search_budget - Annoy param: optional search budget per tree for Annoy. If not provided, it will default to n_tree * k * 20L.

  • diversify_prob - NNDescent param: diversification probability for the NNDescent index. This will diversify the index at the end and identify potentiall better edges. Defaults to 0.0.

  • delta - NNDescent param: early termination criterium for NNDescent. Defaults to 0.001.

  • ef_budget - NNDescent param: optional query budget parameter. Can accelerate querying, but at the cost of Recall.

  • m - HNSW param: number of connections between layers for HNSW. Defaults to 16L.

  • ef_construction - HNSW param: size of dynamic candidate list during construction. Defaults to 200L.

  • ef_search - HNSW param: size of candidate list (higher = better recall, slower). Defaults to 100L.

  • n_list - IVF param: number of clusters/centroids to generate. Defaults to NULL (sqrt(n) n_lists will be generated in this case).

  • n_probe - IVF param: number of clusters/centroids to query Defaults to NULL (sqrt(n_lists) clusters will be queried in this case).