R语言anomalize包多分组变量异常检测分组识别失效问题求助

核心错误原因 #

• 你在分组后调用as_tibble()的操作会直接移除数据的分组属性，所以glimpse()识别不到分组，后续anomalize包的函数也无法按分组处理多个时间序列
• 代码里存在语法错误：time_decompose()后的管道符写漏了>，只有单独的%会导致运行报错

修正后的可运行代码 #

library(dplyr)
library(lubridate)
library(anomalize)

Test.df2 <-
structure(list(Product = c("A1", "A1", "A1", "A1", "B1", "B1", 
"B1", "B1"), Location = c("A", "A", "A", "A", "A", "B", "A", 
"C"), Period = c("Sep 2017", "Oct 2017", "Nov 2017", "Dec 2017", 
"Sep 2017", "Oct 2017", "Nov 2017", "Dec 2017"), Units = c(5.399993, 
9, 10, 9, 5, 1, 3, 1)), row.names = c(NA, 8L), class = "data.frame")

Test.df2$Period<-lubridate::my(Test.df2$Period)
Test.df2[is.na(Test.df2)]<-0
# 不需要额外转tibble，分组属性会自动保留
Test.df2<-Test.df2 %>% group_by(Product,Location)
glimpse(Test.df2)

Test.df2.anomaly<- Test.df2 %>% 
  time_decompose(Units,merge = T,method = 'stl',frequency = "auto",
                 trend = "auto") %>%
  anomalize(remainder,method = 'gesd',alpha = 0.05,max_anoms = 0.2) %>%
  time_recompose()

验证方法 #

运行代码后调用group_keys(Test.df2)即可看到分组信息已被正确识别，异常检测流程可以正常跑完所有分组的时间序列。

内容的提问来源于stack exchange，提问作者Arvind Menon