This repo is an implemetation of the paper Curriculum-Meta Learning for Order-robust Continual Relation Extraction
Because the size of pretrained embedding file is too large to upload, you can download and extract glove.6B.300d.txt file from the Link.
You can use following script to install the dependencies.
pip install -r requirements.txt
To run the model on the FewRel dataset with default arguments, you can simply run the run.py file as following.
python run_fewrel.py
Change arguments in the __main__ method of run.py can lead to different experiment settings.
Following script can run the experiment on the FewRel dataset with setting, which uses cuda device 1, random seed 100, random task spliting, 100 instances for relation, batch size 50 and sequence offset 3.
python run.py --cuda_id=1 --random_seed=100 --task_arrange=random --train_instance_num=100 -batch_size=50 --sequence_index=3
This repo is published for the double-blind review.