Positive/Negative classification on ACP Corpus

Development Environment

• Python 3.6.5
• PyTorch 1.1.0
• scikit-learn 0.20.3
• Pipenv 2018.11.26

To setup environment, run

$ pipenv sync

Development Environment (for BERT)

• Python 3.8
• PyTorch 1.4.0
• Transformers 3.0.4
• scikit-learn

To setup environment, run

$ pip install -r requirements.txt

Models

Random Forest

$ python src/ML/random_forest.py --train_data path/to/train/file --valid_data path/to/valid/file --test_data path/to/test/file

SVM (BoW)

$ python src/ML/svm_bow.py --train_data path/to/train/file --valid_data path/to/valid/file --test_data path/to/test/file

SVM (tf-idf)

$ python src/ML/svm_tfidf.py --train_data path/to/train/file --valid_data path/to/valid/file --test_data path/to/test/file

Neural Network Models (MLP, BiLSTM, BiLSTMAttn, CNN)

• train (MLP)

  $ python src/nn/train.py --model MLP --batch-size 2048 --epochs 20 --save-path result/mlp.pth --device <gpu-id>
  

• test (MLP)

  $ python src/nn/test.py --model MLP --batch-size 2048 --load-path result/mlp.pth --device <gpu-id>
  

BERT

• train and test

   $/src/bert python run.py --do_train --do_eval --output_dir path/to/save
  

Dataset

ACP Corpus: Automatically Constructed Polarity-tagged Corpus

• tagged and splitted data is located at /mnt/hinoki/ueda/shinjin2019/acp-2.0