ReliefE Hyperparameters

ReliefE has multiple hyperparameters, which determine its performance. Let’s discuss the main ones next.

import reliefe

reliefE_instance = reliefe.ReliefE(embedding_based_distances=True,
    verbose=True,
    use_average_neighbour=True,
    determine_k_automatically=True,
    num_iter=[128])

This code snipped initialized ReliefE with all the functionality described in the paper. The parameters are:

Hyperparameter descriptions. Below you can find key hyperparameters.

Hyperparameter	Description	Values
embedding_based_distances	Whether to embed the input and output space if possible (ReliefE). Instances are compared via cosine distance, but the distance for targets needs to be specified (see below).	True/False
use_average_neighbour	Whether to average the neighbors’ embeddings during computation	True/False
determine_k_automatically	Whether to determine the size of neighborhood	True/False
num_iter	Number of iterations	Integer
normalize_descriptive	Normalization of descriptive attributes?	True/False
latent_dimension	Embedding dimension	Integer
mlc_distance	Distance used for comparison in MLC setting	[“f1”,”cosine”,”hyperbolic”,”hamming”,”accuracy”,”subset”]
sparsity_threshold	If number of non-zero elements is larger than this, sparsify.	float between 0 and 1
samples	Number of samples if the number of instances is too large.	Integer

More detailed descriptions can be found in the method description pages in Index.