Unintended-Bias-in-Toxicity-Classification-Kaggle

Kaggle Unintended Bias in Toxicity Classification Challenge: https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification

Current ensemble model ranks 377^th on Private LB with custom Bias AUC score of 0.93832 against Private LB highest of 0.94734

Ensemble Details:

1. BiLSTM model with 5-fold CV trained using FastText + GloVe embeddings (300d each)
2. Fine-tuned BERT baseline

Potential Areas of Improvement:

1. Finetune BERT baseline over entire training set (present limit is 50% due to computational constraints)
2. Try ensembling different BERT baselines (cased / uncased)