R/mADCF.R

In dCovTS: Distance Covariance and Correlation for Time Series Analysis

mADCF <- function(x, lags, unbiased = FALSE, output = TRUE ) {

  if ( !is.matrix(x) )  x <- as.matrix(x)
  if ( (is.data.frame(x) )  |  ( is.matrix(x) ) ) {
    if ( dim(x)[2] == 1 )  stop( 'Only multivariate time series with dimension d>2.' )
    if ( !is.ts(x) )  x <- as.ts(x)
  }
  if ( !is.numeric(x) )  stop( "'x' must be numeric." )
  if ( !all( is.finite(x) ) )  stop( 'Missing or infitive values.' )
  if ( missing(lags) )  stop( "'lags' is missing with no default." )

  dm <- dim(x)
  n <- dm[1]  ;   d <- dm[2]

  if ( length(lags) == 1 ) {
  
    if ( ( lags >= 0 )  &&  ( lags <= (n - 1) ) ) {
      X <- x[(1 + lags):n, ]
      Y <- x[1:(n - lags), ]
    } else if ( ( lags >= (-n + 1) )  &&  ( lags < 0 ) ) {
      X <- x[1:(n + lags), ]
      Y <- x[(1 - lags):n, ]
    } else {
      stop( "lags must be in the range of -(n-1) and (n-1)" )
    }

    cross.adcf <- matrix(NA, d, d)

    if ( unbiased ) {
      for (i in 1:d)  cross.adcf[i, ] <- dcov::mdcor(X[, i], Y, type = "U")
    } else {
      for (i in 1:d)  cross.adcf[i, ] <- dcov::mdcor( X[, i], Y )
      cross.adcf <- sqrt( cross.adcf )
    }

    if ( output ) {
      cat( "Distance Correlation Matrix at lag: ", lags, "\n" )
      print(cross.adcf)
    }

  } else if ( length(lags) > 1 ) {
    len <- length(lags)
    cross.adcf <- array( dim = c(d, d, len))

    for ( j in 1:len ) {
      cross <- matrix(NA, d, d)
      lag <- lags[j]
      if ( ( lag >= 0 )  &&  ( lag <= (n - 1) ) ) {
        X <- x[(1 + lag):n, ]
        Y <- x[1:(n - lag), ]
      } else if ( ( lag >= (-n + 1) )  &&  ( lag < 0 ) ) {
        X <- x[1:(n + lag), ]
        Y <- x[(1 - lag):n, ]
      } else {
        stop( "lags must be in the range of -(n-1) and (n-1)" )
      }
      
      if ( unbiased ) {
        for (i in 1:d)  cross[i, ] <- dcov::mdcor(X[, i], Y, type = "U")
      } else {
        for (i in 1:d)  cross[i, ] <- dcov::mdcor( X[, i], Y )
        cross <- sqrt( cross )
      }
      cross.adcf[, , j] <- cross
    } ## end  for ( i in 1:len ) 

    dimnames(cross.adcf)[[ 3 ]] <- paste( "Distance Correlation Matrix at lag: ", lags) 
    if ( output )  print(cross.adcf)

  } ## end  if ( length(lags) == 1 ) 

  mADCF <- cross.adcf
}












# mADCF <- function (x, lags, unbiased = FALSE, output = TRUE) {
#     if (!is.matrix(x))
#         x <- as.matrix(x)
#     if ((is.data.frame(x)) | (is.matrix(x))) {
#         if (NCOL(x) == 1)
#             stop("Only multivariate time series with dimension d>2")
#         if (!is.ts(x))
#             x <- as.ts(x)
#     }
#     if (!is.numeric(x))
#         stop("'x' must be numeric")
#     if (!all(is.finite(x)))
#         stop("Missing or infitive values")
#     if (missing(lags))
#         stop("'lags' is missing with no default")
#     n <- as.integer(NROW(x))
#     d <- as.integer(NCOL(x))
#     if ((lags >= 0) && (lags <= (n - 1))) {
#         X <- rbind(x[(1 + lags):n, ])
#         Y <- rbind(x[1:(n - lags), ])
#     }
#     else if ((lags >= (-n + 1)) && (lags < 0)) {
#         X <- rbind(x[1:(n + lags), ])
#         Y <- rbind(x[(1 - lags):n, ])
#     }
#     else {
#         stop("lags must be in the range of -(n-1) and (n-1)")
#     }
#     cross.adcf <- matrix(NA, d, d)
#     for (i in 1:d) {
#         for (j in 1:d) {
#             if (unbiased) {
#                 cross.adcf[i, j] <- bcdcor(X[, i], Y[, j])
#             }
#             else {
#                 cross.adcf[i, j] <- dcor(X[, i], Y[, j])
#             }
#         }
#     }
#     if (output) {
#         cat("Distance Correlation Matrix at lag: ", lags,
#             "\n")
#         print(cross.adcf)
#     }
#     mADCF <- cross.adcf
# }