检测异常值的方法有很多,选择哪种方法需要具体问题具体分析,下面罗列一些常用的方法。
Dynamic Time Warping
Dynamic Time Warping
Dynamic Time Warping 的目标是比较两个依赖于时间的序列 $X:=(x_1, x_2, \cdots, x_N), Y:=(y_1, y_2,\cdots, y_M)$,这些序列可以是离散信号(时间序列),或者更一般地,是在等距时间点采样的特征序列。记 $\mathcal{F}$为特征空间,则 $x_n, y_m\in\mathcal{F},n\in[1:N],m\in[1:M]$。为了比较两个不同的特征 $x,y\in\mathcal{F}$,需要引入一个局部花费度量(local cost measure),有时也称为局部距离度量(local distance measure),其定义为
$$c:\mathcal{F}\times\mathcal{F}\to\mathbb{R}_{\ge0}$$
The sincere advice from J.D.Watson's speech in SUSTech
These following advice were taken from J.D.Watson’s speech in SUSTech that I copied from Qzone. (Actually I don’t know who organized these sentences.) Some of opinions seem to be known to everyone, but are easily overlooked. Most of these advice are about study and research, but I think some are also suitable for work. Emmm as for why I am writing in English, just as learning programming: You cannot learn python well if you only reading code without writing. However, no more than five people know the blog, hahaha……