1. Summary

【必写】，推荐使用 grammarly 检查语法问题，尽量参考论文 introduction 的写作方式。需要写出

1. 这篇文章解决了什么问题？
2. 作者使用了什么方法（不用太细节）来解决了这个问题？
3. 你觉得你需要继续去研究哪些概念才会加深你对这篇文章的理解？

This paper compares the deep learning framework for time series prediction, which has performed well in recent years, with the Gradient Boosting Regression Tree (GBRT) model with good input processing. The experimental results show that a simple model can achieve better results by effectively designing the input-output structure, and sometimes even better than the most advanced deep neural network. In this paper, the author transforms the time series prediction task into a window-based regression problem, and does feature engineering on the input and output of the GBRT model, so that the target value of each training window is connected with the external features, and then flattened to form an input example of multi-output GBRT model. Finally, it is pointed out that the well-configured baseline in the field of time series can not be ignored, and it also shows that the window-based processing can be applied to more networks to achieve good results.

2. 你对于论文的思考

需要写出你自己对于论文的思考，例如优缺点，你的takeaways

优点：
1.这篇文章将近年来一些表现良好，且可复现的时间序列深度神经网络进行了总结
2.指明了在时序预测领域，良好的处理往往更能适应实际问题
3.并没有否认深度网络，而是建议需要加强基线的配置

缺点：
1.收录的深度网络模型并不全面，并不包含Transformer类模型
2.GBRT在处理多元问题时，其对输入的处理使得它更适合处理多元问题

3. 其他

【可选】

GBRT的输入处理

实验结果