ranking_embedding_loss_coeff

Description

The coefficient of the embedding loss applied to train ranking-based link prediction model.

• By default, ranking_emb_loss_coeff is set to 0., which means that the model is solely trained to optimize for the final predicted ranking.

For embedding-based ranking predictions, set lp_ranking_emb_loss_coeff to a positive value. This instructs Kumo to explicitly train embeddings to produce high-quality inner-product scores for ranking prediction.

It is only recommended to change this value in case the user is interested in embeddings. In this case, the recommended value of ranking_emb_loss_coeff is 0.2..

Supported Task Types

• Link Prediction