module

​

module: <ranking | embedding> (Optional)

​

Description

By default, Kumo selects the model architecture and designs the hyperparameter search space that best fits your query. However, if you have specific requirements for your use case, you can customize the model architecture to suit your desired output/objective.

The module to choose from:

• embedding: A link prediction GNN model that optimizes link prediction with entity embedding outputs. The ranking scores can be obtained by fast vector calculations (default to inner product) on the embedding space.
• ranking: A link prediction GNN that is optimized for link prediction performance. It produces the ranking scores directly. It can also produce embeddings but their inner product are not perfectly aligned to the ranking.

​

Supported Task Types

• Link Prediction

​

Default Values