/MCMN

Primary LanguagePythonOtherNOASSERTION

Pre-training to Match for Unified Low-shot Relation Extraction

Codes for our ACL 2022 paper: Pre-training to Match for Unified Low-shot Relation Extraction. [Arxiv]

Usage

the data is on FewRel official page https://github.com/thunlp/FewRel, please download and put into ./data directory.

prerequirement

  • Create a Python 3 environment (3.7 or greater), eg using conda create --name MCMN python=3.9
  • Activate the environment: conda activate MCMN
  • Install the dependency packages: pip install -r requirements.txt

FLEX Task Setup

  • Clone the repository: git clone git@github.com:allenai/flex.git
  • enter into flex directory: cd flex
  • Install the package locally with pip install -e .

FLEX data preparation

  • replace the file flex/fewshot/challenges/__init__.py with file __ROOT__/challenge/__init__.py(This step removes other unrelated tasks in FLEX and only keeps FewRel tasks.）
  • Make dataset:
python -c "import fewshot; fewshot.make_challenge('flex');"

FLEX supervised only models:

./test_flex_ft.sh

FLEX pretrain+supervised Test

./test_flex_pt_ft.sh

FewRel None-Of-The-Above Train and Test

For Supervise only model:

  • na rate 0.15: ./test_nota_0.15_ft.sh
  • na rate 0.5: ./test_nota_0.5_ft.sh

For Pretrain + Supervise model:

  • na rate 0.15: ./test_nota_0.15_pt_ft.sh
  • na rate 0.5: ./test_nota_0.5_pt_ft.sh

License

The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for Noncommercial use only. Any commercial use should get formal permission first.