Relation Extraction Task

This is 🌏Earth Coding Lab's repository for the 2nd boostcamp AI Tech competition (2022.11.14 ~ 2022.12.01 19:00).

The competition is on sentence-level Relation Extraction.

• Here is our Wrap-Up Report (link)
• Here is our Presentation (link)

Contributors

Yong Woo Song	Woo Bin Park	Jin Myeong Ahn	Yeong Cheol Kang	Gang Hyeok Lee
Github	Github	Github	Github	Github

Yong Woo Song   :   Result Analysis • Paper Research • Data Augmentation • Model Implementation
Woo Bin Park   :   Loss Function Analysis • Model Ensemble • Model Implementation
Jin Myeong Ahn   :   Code Refactoring • Model Implementation
Yeong Cheol Kang   :   MLFlow Customization • Model Customization • Model Implementation
Gang Hyeok Lee   :   Paper Research • Data Cleaning • Hyperparameter Tuning • Model Implementation

Hardware Used

• NVIDIA TELSA V100
• Ubuntu 18.04

Hardware Used

• NVIDIA TELSA V100
• Ubuntu 18.04

📄 Guideline

Setup

Install all the prerequisites in one go.

make setup

Code formatting & Check lint

make style

Code Testing

make test

Training

python main.py

Inference

python main.py --do_train=False --do_inference
python main.py --do_train=False --do_inference

Run Dashboard

make dashboard

Then you can acess dashboard through your web browser. (If you use VScode, check issue)

---

Data

In this competition, KLUE dataset for relation extraction was used. It can be downloaded from the link below: https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000075/data/dataset.tar.gz

EDA

EDA can be seen from eda_basics.ipynb file under eda folder or check our Wrap-Up Report.

Leaderboard

	Micro F1	AUPRC	Rank
Public	74.5240	76.6242	7th
Private	74.5271	81.1231	2nd