Toxic-Comment-Classification

NOTE: Since I run this code on my local PC, i need to reduce batch_size and max_sequence_length

downloading dataset

mkdir dataset
cd dataset
kaggle competitions download -c jigsaw-toxic-comment-classification-challenge
unzip -q jigsaw-toxic-comment-classification-challenge.zip
unzip -q train.csv.zip

folder structure

• download_tokenizer_model.ipynb: use this one to save pretrained on my pc
• prepare_data.ipynb: use this one to preprocess data and export to a new data
• utils.py: helper function
• train.py: training
• ./bert/: where I saved both tokenizer and model
• dataset/: data here

todo

• add scheduler to optimizer
• train with bigger batch_size, full dataset
• inference