feature-based approach with meanpooling: https://www.kaggle.com/ziliwang/feature-based-approach-with-meanpooling
feature-based approach with attention: https://www.kaggle.com/ziliwang/feature-based-approach-with-attention
fine-turning with attention: https://www.kaggle.com/ziliwang/fine-tuning-with-attention
2019-5-22: fixed bug of tuning embedding layer in BERT
performance changes:
training detail of fine-turning with attention (fixed): https://www.kaggle.com/ziliwang/fork-of-fine-tuning-with-attention