CSE244_ML_for_NLP

Code for CSE244 winter 2020 homework 1: Relation Extraction From Text

Staring script:

run.ipynb or run-bert.ipynb

Report:

https://github.com/MollyZhang/CSE244_ML_for_NLP_HW1/blob/master/HW1%20report.pdf

Folder structure:

• model_utils.py

  This file contains all models I have tried

• data_utils.py

  This contains preparing data with torchtext, for both ngram format and also word embedding format. This file also include wrappers which generate data batches for training.

• train_utils.py

  This file contains the function used for training and calculating losses during training

• evaluation.py

  This file contains calculation of f1 scores in various ways

• submission.py

  This file generate submission file for kaggle. It also includes an Ensemble class to combine results of multiple models into one final result. It also includes basic error anaylysis.

• preprocessing.ipynb

  This is the notebook for preprocessing: train, validation, test splitting. Converting label text to numerical categories. Among with many minute things I tried.

• movie_and_person_names.ipynb

  This is the notebook specificially for one type of feature engineering: processing movie and people names into phrases.

• run.ipynb and run-bert.ipynb

  This should be the starting point. This is where I import all other functions from *.py files and get results.