如何在 R 中使用 Monte Carlo 作为 ARIMA 模拟函数

How to use Monte Carlo for ARIMA Simulation Function in R

这是我想用 R 做的算法:

  1. 通过arima.sim()函数从ARIMA模型模拟10个时间序列数据集
  2. 将系列分成可能的子系列 2s3s4s5s6s7s8s,和 9s
  3. 对于每个大小,通过 auto.arima() 函数对每个块大小的子系列进行重新采样并替换新系列,并从子系列中获得最佳 ARIMA 模型。
  4. 获取每个块大小RMSE的每个子系列。

下面的 R 函数可以完成这项工作。

## Load packages and prepare multicore process
library(forecast)
library(future.apply)
plan(multisession)
library(parallel)
library(foreach)
library(doParallel)
n_cores <- detectCores()
cl <- makeCluster(n_cores)
registerDoParallel(cores = detectCores())
## simulate ARIMA(1,0, 0)
#n=10; phi <- 0.6; order <- c(1, 0, 0)
bootstrap1 <- function(n, phi){
  ts <- arima.sim(n, model = list(ar=phi, order = c(1, 0, 0)), sd = 1)
  ########################################################
  ## create a vector of block sizes
  t <- length(ts)    # the length of the time series
  lb <- seq(n-2)+1   # vector of block sizes to be 1 < l < n (i.e to be between 1 and n exclusively)
  ########################################################
  ## This section create matrix to store block means
  BOOTSTRAP <- matrix(nrow = 1, ncol = length(lb))
  colnames(BOOTSTRAP) <-lb
  ########################################################
  ## This section use foreach function to do detail in the brace
  BOOTSTRAP <- foreach(b = 1:length(lb), .combine = 'cbind') %do%{
    l <- lb[b]# block size at each instance 
    m <- ceiling(t / l)                                 # number of blocks
    blk <- split(ts, rep(1:m, each=l, length.out = t))  # divides the series into blocks
    ######################################################
    res<-sample(blk, replace=T, 10)        # resamples the blocks
    res.unlist <- unlist(res, use.names = FALSE)   # unlist the bootstrap series
    train <- head(res.unlist, round(length(res.unlist) - 10)) # Train set
    test <- tail(res.unlist, length(res.unlist) - length(train)) # Test set
    nfuture <- forecast::forecast(train, model = forecast::auto.arima(train), lambda=0, biasadj=TRUE, h = length(test))$mean        # makes the `forecast of test set
    RMSE <- Metrics::rmse(test, nfuture)      # RETURN RMSE
    BOOTSTRAP[b] <- RMSE
  }
  BOOTSTRAPS <- matrix(BOOTSTRAP, nrow = 1, ncol = length(lb))
  colnames(BOOTSTRAPS) <- lb
  BOOTSTRAPS
  return(list(BOOTSTRAPS))
}

调用函数

bootstrap1(10, 0.6)

我得到以下结果:

##              2        3         4        5        6        7         8         9
##  [1,] 0.8920703 0.703974 0.6990448 0.714255 1.308236 0.809914 0.5315476 0.8175382

我想按时间顺序重复上面的step 1step 4,然后想到了R中的Monte Carlo技术。因此,我加载了它的包和 运行 下面的函数:

param_list=list("n"=10, "phi"=0.6)
library(MonteCarlo)
MC_result<-MonteCarlo(func = bootstrap1, nrep=3, param_list = param_list)

期待以 matrix 形式获得以下结果:

##           [,2]     [,3]      [,4]    [,5]       [,6]      [,7]      [,8]      [,9]
##  [1,] 0.8920703 0.703974  0.6990448 0.714255  1.308236  0.809914  0.5315476 0.8175382
##  [2,] 0.8909836 0.8457537 1.095148  0.8918468 0.8913282 0.7894167 0.8911484 0.8694729
##  [3,] 1.586785  1.224003  1.375026  1.292847  1.437359  1.418744  1.550254  1.30784

但我收到以下错误消息:

Error in MonteCarlo(func = bootstrap1, nrep = 3, param_list = param_list) : func has to return a list with named components. Each component has to be scalar.

我怎样才能找到获得上述所需结果并使结果可重现的方法?

您收到此错误消息是因为 MonteCarlo 希望 bootstrap1() 接受 一个 参数组合进行模拟,并且它仅 returns 每个复制一个 值 (RMSE)。这里不是这种情况,因为块长度 (lb) 由模拟时间序列 (n) within bootstrap1 的长度决定因此您将获得每次调用的 n - 2 块长度的结果。

一个解决方案是将块长度作为参数传递并适当地重写bootstrap1()

library(MonteCarlo)
library(forecast)
library(Metrics)

# parameter grids
n <- 10 # length of time series
lb <- seq(n-2) + 1 # vector of block sizes
phi <- 0.6 # autoregressive parameter
reps <- 3 # monte carlo replications

# simulation function  
bootstrap1 <- function(n, lb, phi) {
    
    #### simulate ####
    ts <- arima.sim(n, model = list(ar = phi, order = c(1, 0, 0)), sd = 1)
    
    #### devide ####
    m <- ceiling(n / lb) # number of blocks
    blk <- split(ts, rep(1:m, each = lb, length.out = n)) # divide into blocks
    #### resample ####
    res <- sample(blk, replace = TRUE, 10)        # resamples the blocks
    res.unlist <- unlist(res, use.names = FALSE)   # unlist the bootstrap series
    #### train, forecast ####
    train <- head(res.unlist, round(length(res.unlist) - 10)) # train set
    test <- tail(res.unlist, length(res.unlist) - length(train)) # test set
    nfuture <- forecast(train, # forecast
                        model = auto.arima(train), 
                        lambda = 0, biasadj = TRUE, h = length(test))$mean    
    ### metric ####
    RMSE <- rmse(test, nfuture) # return RMSE
    return(
      list("RMSE" = RMSE)
    )
}

param_list = list("n" = n, "lb" = lb, "phi" = phi)

到运行模拟,传递参数以及bootstrap1()MonteCarlo()。对于并行执行的模拟,您需要通过 ncpus 设置核心数。 MonteCarlo 包使用 snowFall,所以它应该 运行 on Windows.

请注意,我还设置了 raw = T(否则结果将是所有复制的平均值)。之前设置种子将使结果可重现。

set.seed(123)
MC_result <- MonteCarlo(func = bootstrap1, 
                        nrep = reps,
                        ncpus = parallel::detectCores() - 1,
                        param_list = param_list,
                        export_also = list(
                         "packages" = c("forecast", "Metrics")
                        ),
                        raw = T)

结果是一个数组。我认为最好通过 MakeFrame():

将其转换为 data.frame
Frame <- MakeFrame(MC_result)

虽然很容易得到一个 reps x lb 矩阵:

matrix(Frame$RMSE, ncol = length(lb), dimnames = list(1:reps, lb))