R中時間序列的相關系數的運行窗口

Question

我希望使用R中的運行窗口來計算12個變量的相關系數。

我的數據存儲在具有％m。％d。％Y％H：％M：％S索引的動物園對象中，這12個變量中的每個變量都有1343個觀測值。 我不知道我要使用什么窗口大小，但是可以根據需要更改它。

@Joshua Ulrich在這里發布了如何使用rollapplyr計算滾動相關性，但是此示例只有兩個變量。 由於我有限的R經驗，我不確定如何合並應用族函數之一來對所有12個變量進行關聯。

任何指針將不勝感激

我的數據如下所示：

> head(wideRawXTS)
                    DO0182U09A3 DO0182U09B3 DO0182U09C3 DO0182U21A1 DO0182U21A2 DO0182U21A3 DO0182U21B1 DO0182U21B2 DO0182U21B3
2017-01-20 16:30:00     -101.50     -103.37     -103.86     -104.78     -104.95     -105.33     -102.50      -99.43     -104.05
2017-01-20 16:45:00     -101.32     -102.75     -104.22     -104.51     -103.94     -105.29     -102.82     -101.99     -103.94
2017-01-20 17:00:00     -101.45     -103.30     -103.93     -104.70     -104.82     -105.13     -103.72     -103.95     -104.25
2017-01-20 17:15:00     -100.91      -95.92      -99.22     -103.83     -104.72     -105.19     -103.57     -101.36     -104.09
2017-01-20 17:30:00     -100.91     -103.04     -104.09     -102.15     -104.91     -105.18     -103.88     -104.09     -103.96
2017-01-20 17:45:00     -100.97     -103.67     -104.12     -105.07     -104.23      -97.48     -103.92     -103.89     -104.01
                    DO0182U21C1 DO0182U21C2 DO0182U21C3
2017-01-20 16:30:00     -104.51     -104.42     -105.17
2017-01-20 16:45:00     -104.74     -104.65     -105.25
2017-01-20 17:00:00     -105.02     -105.04     -105.32
2017-01-20 17:15:00     -103.90     -102.95     -105.16
2017-01-20 17:30:00     -104.75     -105.07     -105.23
2017-01-20 17:45:00     -105.08     -105.14     -104.89

#Importing Data
rawDF <- read.csv("RTWP_DO0182_14Day_Raw.csv",
                  header = F,
                  col.names = c("Period Start Time",
                                "PLMNname",
                                "RNCname",
                                "WBTSname",
                                "WBTSID",
                                "WCELname",
                                "WCELID",
                                "Overload",
                                "AverageRTWP",
                                "DC_RTWP_High_PRX",
                                "RTWP_Threshold"),
                  colClasses = c("character",
                                 "NULL", #drops PLMN name
                                 "NULL", #drops RNC name
                                 "character", 
                                 "integer",
                                 "character",
                                 "integer",
                                 "NULL", #drops Overload
                                 "numeric",
                                 "NULL", #drops DC_RTWP_High_PRX
                                 "NULL"), #drops RTWP_Threshold
                  strip.white=TRUE,
                  na.strings = c("NA", 
                                 "NULL"),
                  as.is = T,
                  skip=2)

#convert period.start.time to POSIXlt
rawDF$Period.Start.Time <- as.POSIXlt(strptime(rawDF$Period.Start.Time, 
                                               format="%m.%d.%Y %H:%M:%S"))

#dcast the long data frame to a wide data frame
wideRawDF <- dcast(rawDF, 
                   Period.Start.Time ~ WCELname,
                   value.var = "AverageRTWP")

#assign the date times as rownames for converting to XTS
rownames(wideRawDF) = wideRawDF[[1]]

#drops the duplicate Period Start Time column since date times are rownames
wideRawDF$Period.Start.Time <- NULL

#Convert wideRawDF to XTS time series
wideRawDF <- as.xts(wideRawDF)

#NA values replaced by interpolated values
wideRawDF <- na.approx(wideRawDF, na.rm = FALSE)

#DF is centered by subtracting the column means and scaled by dividing the
#centered columns by their standard deviations
wideRawDFscaled <- scale(wideRawDF, center = TRUE, scale = TRUE)

#window <- 20
#cors <- combn(names(wideRawDFscaled),2,
#              function(p) rollapply(wideRawDFscaled, window ,
#                                    function(x) cor(x[,p[1]],x[,p[2]]),
#                                    by.column = FALSE))
#colnames(cors) <- combn(names(wideRawDFscaled),2,paste,collapse=".")

#Running window of correlation coefficients
Cor <- function(x) {
  corr <- cor(wideRawDFscaled)
  out <- as.data.frame.table(corr)[lower.tri(corr), ]
  with(out, setNames(Freq, paste(Var1, Var2)))
}
slidingCor <- rollapplyr(wideRawDFscaled, 6, Cor, by.column = FALSE)

Answer 1

您可以使用combn將相關性應用於每對列。 調整您引用的答案：

window <- 20
cors <- combn(colnames(wideRawXTS),2,
              function(p) rollapply(wideRawXTS, window  ,
                                    function(x) cor(x[,p[1]],x[,p[2]]), 
                                    by.column=FALSE))
colnames(cors) <- combn(colnames(wideRawXTS),2, paste, collapse=".")

Answer 2

假設在注釋中可重復顯示3列輸入，並使用寬度為3的窗口進行說明，定義相關函數Cor ，該函數接受矩陣並計算其相關矩陣，以提取下三角部分以消除冗余並添加列名。 現在將其與rollapplyr一起rollapplyr ：

library(xts)

Cor <- function(x) {
    corr <- cor(x)
    out <- as.data.frame.table(corr)[lower.tri(corr), ]
    with(out, setNames(Freq, paste(Var1, Var2)))
}
rollapplyr(xx, 3, Cor, by.column = FALSE)

給予：

                    DO0182U09B3.DO0182U09A3 DO0182U09C3.DO0182U09A3 DO0182U09C3.DO0182U09B3
2017-01-20 16:30:00                      NA                      NA                      NA
2017-01-20 16:45:00                      NA                      NA                      NA
2017-01-20 17:00:00                 0.98573                -0.99613                -0.99671
2017-01-20 17:15:00                 0.98629                 0.95983                 0.99297
2017-01-20 17:30:00                 0.52664                 0.47475                 0.99820
2017-01-20 17:45:00                 0.56204                 0.50460                 0.99769

注意：可重復形式的輸入xx為：

xx <- structure(c(-101.5, -101.32, -101.45, -100.91, -100.91, -100.97, 
-103.37, -102.75, -103.3, -95.92, -103.04, -103.67, -103.86, 
-104.22, -103.93, -99.22, -104.09, -104.12), .Dim = c(6L, 3L), .Dimnames = list(
    NULL, c("DO0182U09A3", "DO0182U09B3", "DO0182U09C3")), index = structure(c(1484947800, 
1484948700, 1484949600, 1484950500, 1484951400, 1484952300), tzone = "", tclass = c("POSIXct", 
"POSIXt")), class = c("xts", "zoo"), .indexCLASS = c("POSIXct", 
"POSIXt"), tclass = c("POSIXct", "POSIXt"), .indexTZ = "", tzone = "")