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
python 3+
- nltk
- pandas
- fasttext
- re
- sklearn
- transformers
- datasets
Use the package manager pip to install fasttext.
pip install fasttext
Run in Colab(which can use the free gpu resource)
!pip install git+https://github.com/huggingface/transformers
!pip install datasets
- change the result directory path and run the script in python3
- change the dataset file path and run the script in python3
!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
- WU, Hao
- CHAN, Chun Kit