wuhao050698 / argument-impact-classification

Argument Impact Classification

This project is for MSBD6000H project 1 Argument Impact Classification.

It contains three python file:

• FastText_model.py: FastText model
• run_bert.py: Roberta model
• ensemble.py: Esemble model

Prerequisites

python 3+

For FastText Model

• nltk
• pandas
• fasttext
• re
• sklearn

For Bert Model

• transformers
• datasets

Installation

For FastText Model

Use the package manager pip to install fasttext.

pip install fasttext

For Bert Model

Run in Colab(which can use the free gpu resource)

!pip install git+https://github.com/huggingface/transformers
!pip install datasets

Usage

For Ensemble Model

• change the result directory path and run the script in python3

For FastText Model

• change the dataset file path and run the script in python3

For Bert Model

!python run_bert.py --model_name_or_path roberta-large --use_fast_tokenizer False --train_file 'train.csv' --validation_file 'valid.csv' --do_train --do_eval --do_predict --test_file 'test.csv' --max_seq_length 100 --per_device_train_batch_size 32 --learning_rate 2e-5 --num_train_epochs 3 --output_dir 'roberta'

parameters:

• model_name_or_path: model(BERT/Roberta/Albert...)
• train_file: training file
• validation_file: validation file
• test_file: test file
• output_dir: output direction
• max_seq_length: max sequence length
• per_device_train_batch_size: batch size
• learning_rate: learning rate
• num_train_epochs: epochs

Authors

• WU, Hao
• CHAN, Chun Kit