AcrE

Source codes for our COLING 2020 paper: Knowledge Graph Embedding with Atrous Convolution and Residual Learning.

Overview of AcrE

We denote our model as AcrE (the abbreviation of Atrous Convolution and Residual Embedding). We design two structures to integrate the standard convolution and atrous convolutions together. One is a serial structure as shown in Figure 1 (a), and the other is a parallel structure as shown in Figure 1 (b). Please refer to the paper for details.

Requirements

Compatible with PyTorch 1.0 and Python 3.x.
Dependencies can be installed using requirements.txt.

Datasets

We use FB15k-237,FB15k,DB100K,Kinship,WN18RR,WN18 datasets for evaluation, and they are included in the repo.

Training and evaluate on datasets

Install all the requirements from requirements.txt.
Execute sh preprocess.sh for extracting the datasets and setting up the environment.

The command for training arce and its prompt are below:

data indicates the dataset used for training the model.
gpu is the GPU used for training the model.
name is the provided name of the run which can be later used for restoring the model.

Execute python acre.py --help for listing all the available options.

#Serial fb15k237
python -u acre.py --data FB15k-237 --batch 128 \
--hid_drop 0.5 --feat_drop 0.2 --lr 0.001 --inp_drop 0.3  --gpu 0 --name fb15k_237_s --way s --train_strategy one_to_x

#Serial fb15k
python -u acre.py --data FB15k --batch 256 \
--hid_drop 0.2 --feat_drop 0.2 --lr 0.001 --inp_drop 0.2  --gpu 0 --name fb15k_s --way s --train_strategy one_to_n

#Serial kinship
python -u acre.py --data kinship --batch 128 \
--hid_drop 0.5 --feat_drop 0.5 --lr 0.001 --inp_drop 0.2  --gpu 0 --name kinship_s --way s --train_strategy one_to_n

#Serial WN18RR
python -u acre.py --data WN18RR --batch 256 \
--hid_drop 0.5 --feat_drop 0.1 --lr 0.00125 --inp_drop 0.2  --gpu 0 --name wn18rr_s --way s --train_strategy one_to_n

#Serial WN18
python -u acre.py --data WN18 --batch 256 \
--hid_drop 0.3 --feat_drop 0.3 --lr 0.0012 --inp_drop 0.2  --gpu 0 --name wn18_s --way s  --train_strategy one_to_n

#Serial DB100K
python -u acre.py --data DB100K --batch 256 \
--hid_drop 0.3 --feat_drop 0.2 --lr 0.0012 --inp_drop 0.2  --gpu 0 --name db100k_s --way s --train_strategy one_to_x

#Parallel fb15k237
python -u acre.py --data FB15k-237 --batch 128 \
--hid_drop 0.5 --feat_drop 0.2 --lr 0.001 --inp_drop 0.3  --gpu 0 --name --fb15k_237_p --way p --train_strategy one_to_x

#Parallel fb15k
python -u acre.py --data FB15k --batch 256 \
--hid_drop 0.2 --feat_drop 0.2 --lr 0.001 --inp_drop 0.2  --gpu 0 --name --fb15k_p --way p  --train_strategy one_to_n

#Parallel kinship
python -u acre.py --data kinship --batch 128 \
--hid_drop 0.5 --feat_drop 0.2 --lr 0.0001 --inp_drop 0.3  --gpu 0 --name --kinship_p --way p  --train_strategy one_to_n

#Parallel WN18RR
python -u acre.py --data WN18RR --batch 256 \
--hid_drop 0.5 --feat_drop 0.1 --lr 0.00125 --inp_drop 0.3  --gpu 0 --name --wn18rr_p --way p --train_strategy one_to_x

#Parallel WN18
python -u acre.py --data WN18 --batch 256 \
--hid_drop 0.3 --feat_drop 0.3 --lr 0.0012 --inp_drop 0.2 --gpu 0 --name --wn18_p --way p --train_strategy one_to_x

#Parallel DB100K
python -u acre.py --data DB100K --batch 256 \
--hid_drop 0.3 --feat_drop 0.2 --lr 0.0012 --inp_drop 0.2  --gpu 0 --name --db100k_p --way p --train_strategy one_to_x

Citation

Acknowledgement

Parts of our codes come from InteractE. Thanks for their contributions.

neukg/AcrE