My plan: - First, copy some solution from somewhere (probably Kaggle) and just use it. Basically, make it work. - Second, implement the model from scratch (only using basic libraries including `numpy`).