TuckER

Introduction

[paper]

Title: TuckER: Tensor Factorization for Knowledge Graph Completion

Authors: Ivana Balažević, Carl Allen, Timothy M. Hospedales

Abstract: TuckER is based on Tucker decomposition of the binary tensor representation of knowledge graph triples. It learns entity embeddings, relation embeddings, and a core tensor for scoring triplets. TuckER is a fully expressive model and can represent any binary relation.

Running with hopwise

Model Hyper-Parameters:

  • embedding_size (int) : The embedding size of entities and relations. Defaults to 64.

  • input_dropout (float) : Input dropout rate. Defaults to 0.3.

  • input_dropout1 (float) : First hidden dropout rate. Defaults to 0.4.

  • input_dropout2 (float) : Second hidden dropout rate. Defaults to 0.5.

  • label_smoothing (float) : Label smoothing factor. Defaults to 0.1.

A Running Example:

Write the following code to a python file, such as run.py

from hopwise.quick_start import run_hopwise

run_hopwise(model='TuckER', dataset='ml-100k')

And then:

python run.py

Tuning Hyper Parameters

If you want to use HyperTuning to tune hyper parameters of this model, you can copy the following settings and name it as hyper.test.

learning_rate choice [0.01,0.005,0.001,0.0005,0.0001]
embedding_size choice [32,64,128]
input_dropout choice [0.2,0.3,0.4]
label_smoothing choice [0.0,0.1,0.2]

Note that we just provide these hyper parameter ranges for reference only, and we can not guarantee that they are the optimal range of this model.

Then, with the source code of hopwise (you can download it from GitHub), you can run the run_hyper.py to tuning:

hopwise tune --config_files=[config_files_path] --params_file=hyper.test

For more details about Parameter Tuning, refer to Parameter Tuning.