会议要点

1.用已发表结果所用的数据用我们的模型跑一遍，验证我们建立的pipeline的有效性和正确性
2.先用L1 loss train完模型后，用 pearson coefficient 作为loss去finetune model
3.询问automate的问题，尽量帮助做到automate。

Next Week

1.Use all ground truth (including NH3 loss) to get the result
2.Use the dataset that publication use to run our pipeline for validation.
3.Train the model with L1 loss, then fine tune the model to maximize the Pearson coefficient.
2.Use multiple metric to evaluate our model performance
3.Optimize the other hyper parameter of our model.