r - ggplot中预测数据与实际数据之间的差距

Question

我正在尝试以漂亮的 ggplot 格式绘制一些数据、拟合值和预测，但是当我以我认为应该工作的方式绘制数据时，我发现真实数据和预测之间存在差距。差距是没有意义的，但如果它消失了就好了。

您可以用来重新创建我的问题的一些 R 代码是：

library(xts)
library(tidyverse)
library(forecast)


dates <- seq(as.Date("2016-01-01"), length = 100, by = "days")

realdata <- arima.sim(model = list(ar = 0.7, order = c(1,1,0)), n = 99)

data <- xts(realdata, order.by = dates)

user_arima <- arima(data, order = c(1,1,0))

user_arimaf <- forecast(user_arima)

fits <- xts(user_arimaf$fitted, order.by = dates)

fcastdates <- as.Date(dates[100]) + 1:10

meancast <- xts(user_arimaf$mean[1:10], order.by = fcastdates)

lowercast95 <- xts(user_arimaf$lower[1:10], order.by = fcastdates)
uppercast95 <- xts(user_arimaf$upper[1:10], order.by = fcastdates)

frame <- merge(data, fits, meancast, uppercast95, lowercast95, all = TRUE, fill = NA)


frame <- as.data.frame(frame) %>% 
  mutate(date = as.Date(dates[1] + 0:(109)))

frame %>% 
  ggplot() +
  geom_line(aes(date, data, color = "Data")) +
  geom_line(aes(date, fits, color = "Fitted")) +
  geom_line(aes(date, meancast, color = "Forecast")) +
  geom_ribbon(aes(date, ymin=lowercast95,ymax=uppercast95),alpha=.25) +
  scale_color_manual(values = c(
    'Data' = 'black',
    'Fitted' = 'red',
    'Forecast' = 'darkblue')) +
  labs(color = 'Legend') +
  theme_classic() +
  ylab("some data") +
  xlab("Date") +
  labs(title = "chart showing a gap",  
       subtitle = "Shaded area is the 95% CI from the ARIMA")

图表如下

我知道 ggplot 现在有一个 geom_forecast 但我想按照我的方式构建这个特定的情节。虽然如果没有其他解决方案，那么我将使用 geom_forecast。

Answer 1

缩小差距需要meancast在空白区域的列中提供一个数据点。我想只使用最后一个“真实”数据点的值是有意义的。

# Grab the y-value corresponding to the date just before the gap.
last_data_value = frame[frame$date == as.Date("2016-04-09"), "data"]

# Construct a one-row data.frame.
extra_row = data.frame(data=NA_real_, 
                       fits=NA_real_, 
                       meancast=last_data_value,
                       uppercast95=last_data_value,
                       lowercast95=last_data_value,
                       date=as.Date("2016-04-09"))

# Add extra row to the main data.frame.
frame = rbind(frame, extra_row)