# 06-22

### XGB, LGB on 25% data

### Feature diff
* user_id
* plan price (特價後的價格)
* plan status

### SR-GNN Dense

MRR: 0.38

MRR 0.42 (imp多過一層dense)

> pytorch cuda error
> https://blog.csdn.net/ShuqiaoS/article/details/88978986
> 如果給loss func的label超出範圍也會導致此錯誤