Few-Shot Link Prediction for Temporal Knowledge Graphs
This repository contains the implementation of the TFSC architectures described in the paper.
Install Pytorch (>= 1.1.0)
pip install pytorch
Python 3.x (tested on Python 3.6)
pip install python 3.6
Numpy
pip install numpy
Pandas
pip install pandas
tqdm
pip install tqdm
run the code:
python train.py --parameters
To run our code, we need to divide the data set according to the data set preprocess file first, or divide it according to our own needs. If we want to get the best results, we need to use Complex to pre-train and then embed it into the model.## Baselines
We use the following public codes for baselines and hyperparameters.
| Baselines | Code | parameters |
|---|---|---|
| TransE | Link | { lr=0.0001, dim=512,b=512} |
| TTransE | link | { lr=0.001, dim=512,b=512} |
| TA-DistMult | link | { lr=0.001, dim=512,b=1024} |
| MateR | [link] | |
| FSRL | [link] | |
| FAAN | [link] | |
| FTMF | [link] |
We implemented DistMult refer to [RotatE](: https://github.com/DeepGraphLearning/ KnowledgeGraphEmbedding.). The user can run the baselines by the following command.
cd ./baselines
bash run.sh train MODEL_NAME DATA_NAME 0 0 512 1024 512 200.0 0.0005 10000 8 0