R/sparse_ar.R

In chaosreader/stat_6550_group_project: This Package Prepares Data and Provides Analysis Tools for Non-Linear Time Series

#' Sparse AR Model with Auto-Selected Coefficients
#' @name sparse_ar
#' @param x A time series object
#' @param p_max The maximum number of coefficients to be compute
#' @param pct_ci What percentage of confidence interval to be used for
#'   coefficient consideration.
#' @param margin An additional margin to increase sparsity.
#' @examples
#' data <- base_impute(bee.data[1:100], impute_col = "TOTAL_COUNT")
#' x <- data$IMPUTED_VALS
#' model <- sparse_ma(x)
#' @export
sparse_ar <- function(x, p_max = 30, pct_ci = .95, margin = 0) {
  n <- length(x)
  d <- seq(1, n)
  se <- stats::qnorm(1 - (1 - pct_ci) / 2)

  criticals <- se / sqrt(n - d)
  criticals <- criticals[1:p_max]


  init_acf <- stats::pacf(x, lag.max = p_max, plot = FALSE)$acf[1:p_max]
  coef_locs <- which(init_acf > criticals * (1 + margin))

  fixed <- rep(0, p_max)
  fixed[coef_locs] <- NA

  model <- stats::arima(x - mean(x),
    order = c(p_max, 0, 0),
    include.mean = FALSE, fixed = fixed,
    transform.pars = FALSE, method = "CSS-ML"
  )

  print(stats::Box.test(model$residuals, lag = 36, type = "Ljung-Box"))

  return(model)
}