基于 Pytorch，使用改良的 Transformer 模型应用于多维时间序列的分类任务上
实验结果
对比模型选择 Fully Convolutional Networks (FCN) and Residual Net works
(ResNet)
引用：[Wang et al., 2017] Z. Wang, W. Yan, and T. Oates. Time series
classification from scratch with deep neural networks:A strong baseline. In
2017 International Joint Conference on Neural Networks (IJCNN), pages 1578–
1585, 2017.
DataSet
FCN
ResNet
Gated Transformer
ArabicDigits
99.4
99.6
98.8
AUSLAN
97.5
97.4
97.5
CharacterTrajectories
99.0
99.0
97.0
CMUsubject16
100
99.7
100
ECG
87.2
86.7
91.0
JapaneseVowels
99.3
99.2
98.7
Libras
96.4
95.4
88.9
UWave
93.4
92.6
91.0
KickvsPunch
54.0
51.0
90.0
NetFlow
89.1
62.7
100
PEMS
-
-
93.6
Wafer
98.2
98.9
99.1
WalkvsRun
100
100
100
