Kehan Yin

hello @ncomly-nvidia, I am a student interested in the project! I want to ask if there are any good-first-issue feature request for Features & Optimizations recently? 🤣

the same problem I got

好的好的！谢谢！我先详细看看，有问题再和您交流！

@WenWeiTHU hello 同学你好，这个例子的确符合您提到的观点，我大概理解你解耦的想法，不过我在今年的NIPS2024看到有一篇**Frequency Adaptive Normalization For Non-stationary Time Series Forecasting**[[arvix]](https://arxiv.org/pdf/2409.20371)上似乎有一个和您思路有些对立的思想，他认为这些高幅值的topk频率反而是不稳定的，文章的introduction里面有简单提到和koopa的比较区别，我还是这方面的初学研究者，想请你指点一下这两个文章都在解决Non-stationary的时序问题，是什么场景和情况导致的出现这两种不同的观点？因为从我的实验经验来看，这两种观点有时候确实都有可用之处🤣，其实从他给出的证明角度有点不太能说服我，因为他说去掉频率之后方差会变小我能理解，但是不论是去掉哪个部分都是会减小方差，这和稳不稳定似乎关系不大。  

Hello @li-jin-gou, I am interested in this request. I want to ask DDL-Time for the PR and if there are any open-source libraries and documentation I can follow. It may...

the reason come from the Model used for LSA64，should I ignore the require_grad = False parameters? I try to skip these layers in ANDMask grad computing it does works now,...

Hi, @horral . Thank you for pointing out the issue. This problem has been fixed in latest dynamicgo version. PR: https://github.com/cloudwego/dynamicgo/pull/99