达摩院提出时序预测新模型FEDformer

顶会点赞！达摩院提出时序预测新模型

阿里云 2022-07-12 16:05

https://mp.weixin.qq.com/s/9doHueBCbsV7eUH2q3uv0A

代码：https://github.com/DAMO-DI-ML/ICML2022-FEDformer
论文：https://arxiv.org/abs/2201.12740

框架：

任务和数据集：

数据集下载地址：https://cloud.tsinghua.edu.cn/d/e1ccfff39ad541908bae/

结论：

环境：Python 3.6, PyTorch 1.9.0

依赖：

pandas
numpy
torch
sklearn
einops
sympy
matplotlib

log.run_S:

>>>>>>>start training : ETTh1_FEDformer_random_modes64_ETTh1_ftS_sl96_ll48_pl96_dm512_nh8_el2_dl1_df2048_fc3_ebtimeF_dtTrue_Exp_2>>>>>>>>>>>>>>>>>>>>>>>>>>
train 8449
val 2785
test 2785
Epoch: 1 cost time: 69.89365196228027
Epoch: 1, Steps: 264 | Train Loss: 0.2078484 Vali Loss: 0.1244287 Test Loss: 0.0884568
Validation loss decreased (inf --> 0.124429).  Saving model ...
Updating learning rate to 0.0001
Epoch: 2 cost time: 70.45653772354126
Epoch: 2, Steps: 264 | Train Loss: 0.1732078 Vali Loss: 0.1181202 Test Loss: 0.0896501
Validation loss decreased (0.124429 --> 0.118120).  Saving model ...
Updating learning rate to 5e-05

log.run_M:

>>>>>>>start training : ETTm1_FEDformer_random_modes64_ETTm1_ftM_sl96_ll48_pl96_dm512_nh8_el2_dl1_df2048_fc3_ebtimeF_dtTrue_Exp_0>>>>>>>>>>>>>>>>>>>>>>>>>>
train 34369
val 11425
test 11425
Epoch: 1 cost time: 4989.59953212738
Epoch: 1, Steps: 1074 | Train Loss: 0.3408678 Vali Loss: 0.4292768 Test Loss: 0.3639539
Validation loss decreased (inf --> 0.429277).  Saving model ...
Updating learning rate to 0.0001

标签：Loss,FEDformer,时序,Test,Epoch,https,达摩院,...
来源： https://www.cnblogs.com/xuehuiping/p/16471866.html