Serializing Knowledge Graphs

[Demirrr]

While we are using our models on DBpedia knowledge graph containing 10^8 triples, we observe serialization and deserialization of large data can be useful.

1. Serialize and Deserialize (KG object).
2. Use the path of serialized object as a parameters so that we could avoid reparsing large input file, if

• a) during computation exception occurs.
• b) we want to train a different kge model on the same dataset.
• c) we want to train the kge model with different hypeparam configuration.

Number of triples: 375_434_888 Number of entities: 114747963 Number of relations: 13906

[Demirrr]

By using --deserialize_flag, we are able to deserialized a KG. This allow us to avoid reparsing large KGs.