ranking_embedding_loss_coeff

Description

A list of coefficients of the embedding loss applied to train ranking-based link prediction model for AutoML to explore.

By default, ranking_embedding_loss_coeff is set to 0.0, which means that the model is solely trained to optimize for the final predicted ranking.

It is only recommended to change this value in case you are interested in embeddings. For embedding-based ranking predictions, the recommended value of ranking_embedding_loss_coeff is 0.2, or a positive value otherwise. This instructs Kumo to explicitly train embeddings to produce high-quality inner-product scores for ranking prediction.

Supported Task Types

• Temporal Link Prediction

Default Values