HITsz-TMG/PPAT

PPAT: Progressive Graph Pairwise Attention Network for Event Causality Identification

PPAT: Progressive Graph Pairwise Attention Network for Event Causality Identification

Setup

conda create -n PPAT python=3.8
conda activate PPAT
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt

Download BERT-base from here and put it under encoder/BERT-base/ folder. Editing the vocab.txt as following (adding <t> and </t> as event marker tokens):

[PAD]
<t>
</t>
[unused2]
[unused3]
[unused4]
[unused5]
...

Data

We use EventStoryLine(v0.9) and Causal-TimeBank. The data is in the ''data'' folder. The MAVEN-ERE data set can be found in here

Running

running cross-validation on EventStoryLine

cd PPAT
python PPAT_framework.py --device {GPU device} --run_mode stack5

running one fold on EventStoryLine

cd PPAT
python PPAT_framework.py --device {GPU device} --run_mode train0

running one fold on Causal-TimeBank

cd PPAT
python PPAT_framework.py --device {GPU device} --run_mode train0 --stack_datapath "../data/ctb_stack10_123"  --dev_datapatg "../data/ctb_stack10_123/0/test.json"