问题

以下是翻译好的部分：

我正在尝试拟合一个一步预测的时间序列模型，所以我需要在每次迭代中扩展我的训练集，如下所示：

library(forecast)
(AirPassengers)
in.sample <- window(AirPassengers, c(1949,1), c(1958,12))
out.sample <- window(AirPassengers, c(1959,1), c(1960,12))
# initialize
hw_forecast <- numeric(); day_fcast <- character()
class(day_fcast) <- "Date"
 
for (i in 1:length(out.sample)) {
         in.sample <- window(AirPassengers, start = c(1991, 1), 
                              end = as.Date(time(out.sample[i]))-1)
         fit= hw(train.ts, seasonal="additive")
         fore1 <- forecast(fit, h=1)
         hw_forecast[i] <- fore1$mean # fore1[[2]][[1]]
         day_fcast[i] <- as.Date(time(out.sample[i,]))
     }

然而，报告了以下错误：

Error in window.default(x, ...) : 'start' cannot be after 'end'

如果您想知道如何确定窗口的结束日期，请告诉我。

library(forecast)
(AirPassengers)
in.sample  <- window(AirPassengers, c(1949,1), c(1958,12))
out.sample <- window(AirPassengers, c(1959,1), c(1960,12))
# initialize
hw_forecast <- numeric(); day_fcast <- character()
class(day_fcast) <- "Date"
 
for (i in 1:length(out.sample)) {
         in.sample <- window(AirPassengers, start = c(1991, 1), 
                              end = as.Date(time(out.sample[i]))-1)
         fit= hw(train.ts, seasonal="additive")
         fore1 <- forecast(fit, h=1)
         hw_forecast[i] <- fore1$mean # fore1[[2]][[1]]
         day_fcast[i] <- as.Date(time(out.sample[i,]))
     }

Error in window.default(x, ...) : 'start' cannot be after 'end'

答案1

得分: 1

以下是要翻译的代码部分：

使用`subset()`函数来解决这个问题更简单。

```{r}
library(forecast)

in.sample  <- window(AirPassengers, c(1949,1), c(1958,12))
out.sample <- window(AirPassengers, c(1959,1), c(1960,12))
# 初始化
hw_forecast <- out.sample*NA

for (i in seq_along(out.sample)) {
  train.ts <- subset(AirPassengers, end = length(in.sample) + i)
  fit <- hw(train.ts, seasonal="additive")
  fore1 <- forecast(fit, h=1)
  hw_forecast[i] <- fore1$mean
}

• 在你的代码中未定义train.ts。我假设你指的是在前一行创建的ts对象。
• 这些是月度数据，因此使用as.Date()没有意义。相反，我已经将hw_forecast设置为具有正确起始和频率属性的ts对象。
• as.Date()不能用于设置ts对象的frequency，因为frequency应该是数值型的。

<details>
<summary>英文:</summary>

It is simpler to use the `subset()` function for this problem.

```{r}
library(forecast)

in.sample  &lt;- window(AirPassengers, c(1949,1), c(1958,12))
out.sample &lt;- window(AirPassengers, c(1959,1), c(1960,12))
# initialize
hw_forecast &lt;- out.sample*NA

for (i in seq_along(out.sample)) {
  train.ts &lt;- subset(AirPassengers, end = length(in.sample) + i)
  fit &lt;- hw(train.ts, seasonal=&quot;additive&quot;)
  fore1 &lt;- forecast(fit, h=1)
  hw_forecast[i] &lt;- fore1$mean
}

• train.ts is undefined in your code. I have assumed you mean the ts object created on the previous line.
• These are monthly data, so it makes no sense to use as.Date(). Instead, I have made hw_forecast a ts object with the correct start and frequency attributes.
• as.Date() can't be used to set the frequency of a ts object, as the frequency should be numeric.