ItoE (Knowledge Graph Embeddings using Neural It ˆ𝑜 Process: From Multiple Walks to Stochastic Trajectories)

Installation:

Install anaconda and required environment (https://docs.anaconda.com/anaconda/install/index.html)

Run the following command to set the environment

conda env create -f environment.yml

Activate the conda environment

conda activate code_env

Set the environment variables

source set_env.sh

Datasets

Download and pre-process the datasets:

source datasets/download.sh
python datasets/process.py

Models

The framework includes:

• TransE
• DistMult
• RotatE
• ComplEx
• RotE
• RotH
• ATTE
• ATTH
• REFE
• REFH
• ItoE (Our model - Euclidean) - Currently under the name SDE in the script
• ItoP (Our model - Poincare) - Currently under the name SDP in the script

Training

Run the example commands from the example commands folder

./train_SDP_WN18RR_32.sh

NB: One needs to make coomands executable before ruunning

chmod +x train_SDP_WN18RR_32.sh

Alternatively, one can run following example command (it is the optimal command for WN18RR for reproducing the results for this dataset, the best results mentioned in the paper was obtained based on early stopping criteria):

python  run.py \
            --dataset WN18RR \
            --model SDP \
            --rank 32 \
            --regularizer N3 \
            --reg 0.0 \
            --optimizer Adam \
            --max_epochs 400 \
            --patience 20 \
            --valid 5 \
            --batch_size 100 \
            --neg_sample_size 500 \
            --init_size 0.001 \
            --learning_rate 0.001 \
            --gamma 0.0 \
            --bias learn \
            --dtype single \
            --double_neg \
            --multi_c \
            --cuda_n 0

Citation

This code is based on the implementation of the following paper:

@inproceedings{chami2020low,
  title={Low-Dimensional Hyperbolic Knowledge Graph Embeddings},
  author={Chami, Ines and Wolf, Adva and Juan, Da-Cheng and Sala, Frederic and Ravi, Sujith and R{\'e}, Christopher},
  booktitle={Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  pages={6901--6914},
  year={2020}
}