deepkashiwa20 / DL-Traff-Grid

[CIKM 2021 Resource Paper] DL-Traff: Survey and Benchmark of Deep Learning Models for Urban Traffic Prediction (Grid Part)
50 stars 15 forks source link

期待更强的模型 #1

Open kpmokpmo opened 3 years ago

kpmokpmo commented 3 years ago

非常感谢你们的工作,很系统并有指导意义,相信持续更新可以做成相关领域的重要指南。根据我个人的经验有几个模型,可能会有更亮眼表现。

  1. StemGNN: Spectral Temporal Graph Neural Network for Multivariate Time-series Forecasting
  2. TS2VEC:Learning Timestamp-Level Representations for Time Series with Hierarchical Contrastive Loss
  3. Query Selector:LONG-TERM SERIES FORECASTING WITH QUERY SELECTOR – EFFICIENT MODEL OF SPARSE ATTENTION

它们主要适用于Graph类的数据集,stemgnn在我的私有数据集上显著优于MTGNN。TS2VEC是multivariates的非监督型的模型,需要调参。Query Selector只是sparse版的transformer,但它自己列的表现较为优秀,优于Informer。如果您对以上三个可能SOTA的候选感兴趣的话不妨实验一下,如果没精力就算了哈:p。

感谢!

deepkashiwa20 commented 3 years ago

好的 感谢感谢

最近时序列这里确实有很多好模型不断涌现,我们小团队自己人数有限,没法能覆盖特别多。

这三篇文章我们会尽快开始加入我们的benchmark,感谢您的支持。

我们这篇工作有幸被CIKM 2021 Resource Track 录用了,

预印版放到了arXiv上面,里面主要整理了一些交通这边的时序列工作,希望对您的研究有帮助。

https://arxiv.org/pdf/2108.09091.pdf https://arxiv.org/pdf/2108.09091.pdf%0d

万分感谢!

From: kpmokpmo @.> Sent: Monday, August 23, 2021 6:38 PM To: deepkashiwa20/DL-Traff-Grid @.> Cc: Subscribed @.***> Subject: [deepkashiwa20/DL-Traff-Grid] 期待更强的模型 (#1)

非常感谢你们的工作,很系统并有指导意义,相信持续更新可以做成相关领域的重要指南。根据我个人的经验有几个模型,可能会有更亮眼表现。

  1. StemGNN: Spectral Temporal Graph Neural Network for Multivariate Time-series Forecasting
  2. TS2VEC:Learning Timestamp-Level Representations for Time Series with Hierarchical Contrastive Loss
  3. Query Selector:LONG-TERM SERIES FORECASTING WITH QUERY SELECTOR – EFFICIENT MODEL OF SPARSE ATTENTION

它们主要适用于Graph类的数据集,stemgnn在我的私有数据集上显著优于MTGNN。TS2VEC是multivariates的非监督型的模型,需要调参。Query Selector只是sparse版的transformer,但它自己列的表现较为优秀,优于Informer。如果您对以上三个可能SOTA的候选感兴趣的话不妨实验一下,如果没精力就算了哈:p。

感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/deepkashiwa20/DL-Traff-Grid/issues/1 , or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5RDG7LMWYDTU56GGZUC3TT6IJG3ANCNFSM5CUF7BHQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email . https://github.com/notifications/beacon/AM5RDG3Z2NAJ7CYCPQJD7T3T6IJG3A5CNFSM5CUF7BH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4ORZJKKA.gif

kpmokpmo commented 3 years ago

谢谢,会持续关注你们的工作,已经watch & star