Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



15 Commits

Repository files navigation

RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space


This is the PyTorch implementation of the RotatE model for knowledge graph embedding (KGE). We provide a toolkit that gives state-of-the-art performance of several popular KGE models. The toolkit is quite efficient, which is able to train a large KGE model within a few hours on a single GPU.

A faster multi-GPU implementation of RotatE and other KGE models is available in GraphVite.

Implemented features


  • RotatE
  • pRotatE
  • TransE
  • ComplEx
  • DistMult

Evaluation Metrics:

  • MRR, MR, HITS@1, HITS@3, HITS@10 (filtered)
  • AUC-PR (for Countries data sets)

Loss Function:

  • Uniform Negative Sampling
  • Self-Adversarial Negative Sampling


Knowledge Graph Data:

  • entities.dict: a dictionary map entities to unique ids
  • relations.dict: a dictionary map relations to unique ids
  • train.txt: the KGE model is trained to fit this data set
  • valid.txt: create a blank file if no validation data is available
  • test.txt: the KGE model is evaluated on this data set


For example, this command train a RotatE model on FB15k dataset with GPU 0.

CUDA_VISIBLE_DEVICES=0 python -u codes/ --do_train \
 --cuda \
 --do_valid \
 --do_test \
 --data_path data/FB15k \
 --model RotatE \
 -n 256 -b 1024 -d 1000 \
 -g 24.0 -a 1.0 -adv \
 -lr 0.0001 --max_steps 150000 \
 -save models/RotatE_FB15k_0 --test_batch_size 16 -de

Check argparse configuration at codes/ for more arguments and more details.


CUDA_VISIBLE_DEVICES=$GPU_DEVICE python -u $CODE_PATH/ --do_test --cuda -init $SAVE

Reproducing the best results

To reprocude the results in the ICLR 2019 paper RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space, you can run the bash commands in to get the best performance of RotatE, TransE, and ComplEx on five widely used datasets (FB15k, FB15k-237, wn18, wn18rr, Countries).

The script provides an easy way to search hyper-parameters:

bash train RotatE FB15k 0 0 1024 256 1000 24.0 1.0 0.0001 200000 16 -de


The KGE models usually take about half an hour to run 10000 steps on a single GeForce GTX 1080 Ti GPU with default configuration. And these models need different max_steps to converge on different data sets:

Dataset FB15k FB15k-237 wn18 wn18rr Countries S*
MAX_STEPS 150000 100000 80000 80000 40000
TIME 9 h 6 h 4 h 4 h 2 h

Results of the RotatE model

Dataset FB15k FB15k-237 wn18 wn18rr
MRR .797 ± .001 .337 ± .001 .949 ± .000 .477 ± .001
MR 40 177 309 3340
HITS@1 .746 .241 .944 .428
HITS@3 .830 .375 .952 .492
HITS@10 .884 .533 .959 .571

Using the library

The python libarary is organized around 3 objects:

  • TrainDataset ( prepare data stream for training
  • TestDataSet ( prepare data stream for evluation
  • KGEModel ( calculate triple score and provide train/test API

The file contains the main function, which parses arguments, reads data, initilize the model and provides the training loop.

Add your own model to like:

def TransE(self, head, relation, tail, mode):
    if mode == 'head-batch':
        score = head + (relation - tail)
        score = (head + relation) - tail

    score = self.gamma.item() - torch.norm(score, p=1, dim=2)
    return score


If you use the codes, please cite the following paper:

 title={RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space},
 author={Zhiqing Sun and Zhi-Hong Deng and Jian-Yun Nie and Jian Tang},
 booktitle={International Conference on Learning Representations},


No description, website, or topics provided.







No releases published


No packages published

Contributors 4