This is the codebase to run the experiments described in paper 'Improving Knowledge Graph Embeddings with Inferred Entity Types' train.py provides a command-line interface to run the experiments. test.py runs 2 training epochs on various combinations of the parameters to check that the code runs.
Main files are in src/: experiment.py - Loads the dataset and constructs the data structures for running the experiments. modules.py - Contains several classes for models used in experiment. Joint_Module is the module for joint training, and needs to be passed submodules. trainer.py - Contains the Trainer object that is responsible for training the modules. batching.py - Responsible for generating batches and negative examples to feed to the Trainer
datasets can be obtained from: FB15k-237 and YAGO3-10: https://github.com/TimDettmers/ConvE FB15K: https://everest.hds.utc.fr/doku.php?id=en:transe
This library is licensed under the Apache 2.0 License.