预测::thetaf 错误答案

【问题标题】：forecast::thetaf error预测::thetaf 错误
【发布时间】：2018-03-02 10:15:54
【问题描述】：

我目前在使用预测包 8.2 和 thetaf 功能时遇到问题：

dat<- structure(c(5, 0, 5, 0, 0, 2, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.9, 0, 2, 0, 1, 
0, 2.1, 0, 2, 0, 1, 0, 0, 0, 2.5, 2, 2, 0, 1.7, 0, 1.5, 0, 1, 
0, 0, 0, 2.5, 0), .Tsp = c(1999, 2003.91666666667, 12), class = "ts")
library(forecast)
thetaf(dat,h = 1)$mean

出现以下错误

Error in ets(object, lambda = lambda, allow.multiplicative.trend =  allow.multiplicative.trend,  : 
  y should be a univariate time series
Additional: Warnings:
1: In ets(x, "ANN", alpha = alpha, opt.crit = "mse", lambda = lambda,  :
  Missing values encountered. Using longest contiguous portion of time series
2: In fit$call <- match.call() : ...

错误与频率有关。如果它是 1 而不是 12，那么它可以工作。我看不出这会导致错误的原因。有任何想法吗？

【问题讨论】：

如果零是问题，如果我将频率降低到 1，为什么该函数会起作用？

标签： r time-series forecasting

【解决方案1】：

当应用于季节性数据时，theta 方法使用乘法经典分解来消除季节性。在这种情况下，分解失败：

> decompose(dat, type="multiplicative")
$x
     Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1999 5.0 0.0 5.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0 0.0 0.0
2000 4.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
2001 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.9 0.0 2.0 0.0
2002 1.0 0.0 2.1 0.0 2.0 0.0 1.0 0.0 0.0 0.0 2.5 2.0
2003 2.0 0.0 1.7 0.0 1.5 0.0 1.0 0.0 0.0 0.0 2.5 0.0

$seasonal
           Jan       Feb       Mar       Apr       May       Jun       Jul       Aug
1999 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
2000 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
2001 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
2002 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
2003 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
           Sep       Oct       Nov       Dec
1999 0.7674245 0.0000000 2.1696136 0.7889410
2000 0.7674245 0.0000000 2.1696136 0.7889410
2001 0.7674245 0.0000000 2.1696136 0.7889410
2002 0.7674245 0.0000000 2.1696136 0.7889410
2003 0.7674245 0.0000000 2.1696136 0.7889410

$trend
           Jan       Feb       Mar       Apr       May       Jun       Jul       Aug
1999        NA        NA        NA        NA        NA        NA 0.9583333 0.9166667
2000 0.3333333 0.3333333 0.3333333 0.3333333 0.3333333 0.3333333 0.1666667 0.0000000
2001 0.0000000 0.0000000 0.0375000 0.0750000 0.1583333 0.2416667 0.2833333 0.3250000
2002 0.7083333 0.7500000 0.7125000 0.6750000 0.6958333 0.8000000 0.9250000 0.9666667
2003 0.8916667 0.8916667 0.8916667 0.8916667 0.8916667 0.8083333        NA        NA
           Sep       Oct       Nov       Dec
1999 0.7083333 0.5000000 0.5000000 0.4166667
2000 0.0000000 0.0000000 0.0000000 0.0000000
2001 0.4125000 0.5000000 0.5833333 0.6666667
2002 0.9500000 0.9333333 0.9125000 0.8916667
2003        NA        NA        NA        NA

$random
           Jan       Feb       Mar       Apr       May       Jun       Jul       Aug
1999        NA       NaN        NA       NaN        NA        NA 0.0000000       NaN
2000 2.1793043       NaN 0.0000000       NaN 0.0000000       NaN 0.0000000       NaN
2001       NaN       NaN 0.0000000       NaN 0.0000000       NaN 0.0000000       NaN
2002 0.2563887       NaN 2.3017827       NaN 2.3911982       NaN 3.7907196       NaN
2003 0.4073466       NaN 1.4889369       NaN 1.3995214       NaN        NA       NaN
           Sep       Oct       Nov       Dec
1999 0.0000000       NaN 0.0000000 0.0000000
2000       NaN       NaN       NaN       NaN
2001 2.8430397       NaN 1.5802682 0.0000000
2002 0.0000000       NaN 1.2627714 2.8430397
2003        NA       NaN        NA        NA

$figure
 [1] 5.5063443 0.0000000 1.2804721 0.0000000 1.2020131 0.0000000 0.2851915 0.0000000
 [9] 0.7674245 0.0000000 2.1696136 0.7889410

$type
[1] "multiplicative"

attr(,"class")
[1] "decomposed.ts"

出现问题是因为 4 月、6 月、8 月和 10 月的季节性指数为零，因此除以季节性指数会产生 NaN 值。

我已经解决了这个问题（在预测包的 v8.3 中），以便在发生这种情况时继续使用非季节性版本的 theta 方法。

【讨论】：