Sprint #14 (Aug 3)

[pagand]

• Github: Update git project regularly and push the finished part for review
• Model config
  • Mode as action vs mode as a dependent feature:
  • If the model can predict Mode with high accuracy, leave it as a dependent feature, otherwise consider mode as a action
  • If mode as action, might consider add penalty for frequent mode change
• Gym environment performance optimization
  • Step() method needs to be fast, move inverse transform to render()
  • Put predict() method under the Gym class
• Optimization model:
  • Offline rl model, use gym for evaluation/test only
• Paper:
  • AAAI conference proposal paper
  • 2-page abstract, deadline Sep. 15
  • Finish by the end of 3rd week in August

[pagand]

@yim-fan Any update from this week? You have not yet accepted the overleaf any problem?? Please update the status in Github

[yim-fan]

@pagand This week I am working on two main things:

1. the optimization, which includes some experiments and literature reviews
2. Improve the FC model by using cox-box scaler or other power scaler to transform some of the skewed features such as speed and fuel consumption. I will update the model code once this part is done.