kumoai.trainer.NeighborSamplingPlan

kumoai.trainer.NeighborSamplingPlan#

class kumoai.trainer.NeighborSamplingPlan[source]#

Configuration parameters that define how subgraphs are sampled in the training and batch prediction pipelines. Please see the Kumo documentation for more information.

Variables:

  • num_neighbors – (list[NumNeighborsConfig]) Determines the number of neighbors to sample for each hop when sampling subgraphs for training and prediction (default: run_mode-dependent). Supported Task Types: All

  • sample_from_entity_table – (bool) Whether to include the entity table in sampling (default: True). Supported Task Types: Static

  • adaptive_sampling – (bool) Whether to use the MetapathAware adaptive sampling algorithm. If True, sampler will oversample nodes in later hops whenever it samples fewer than the specified number of neighbors in a hop. Generally improves performance at the cost of a longer runtime. (default: False). Supported Task Types: All

Contents