DS-GA-1003 project
- Draw games are seldom predicted correctly (almost 100 percent wrong)
- Model seldom predicts a outcome to be draw
- Beta feature rectify one game prediction to be correct, reason?
- Consider the teams occupied in each wrong prediction
- Add player information, e.g. which players have played in a given game, might use embedding roster (like CBOW)
- for nonlinear model, just use a bit value to label teams is fine