在R中如何将数据框/数值对象转为forecast对象以适配autoplot？

Got it, let's tackle this problem! The error you're seeing happens because autoplot() from the forecast package only works with objects of class forecast—and Prophet's predictions come as a data frame (or numeric vectors like yhat), which don't fit that mold. Here's how to convert your Prophet forecast into a forecast object that plays nicely with autoplot():

Step 1: Understand the forecast Object Structure #

A valid forecast object is a list with specific required elements:

• x: The historical time series (as a ts object)
• mean: The predicted values (as a ts object)
• lower/upper: Prediction intervals (as matrix-style ts objects)
• It also needs the forecast class assigned.

Step 2: Full Conversion Example #

Let's walk through a complete example using sample data (you can swap this with your own dataset):

# Load required libraries
library(forecast)
library(prophet)
library(dplyr)

# 1. Prepare your time series data (using AirPassengers as an example)
df <- data.frame(
  ds = as.Date(time(AirPassengers)),
  y = as.numeric(AirPassengers)
)

# 2. Run Prophet model and generate forecasts
m <- prophet(df)
future <- make_future_dataframe(m, periods = 12, freq = "month")
prophet_forecast <- predict(m, future)

# 3. Convert historical data to a ts object
# Adjust `frequency` and `start` to match your data (e.g., 12 for monthly, 365 for daily)
historical_ts <- ts(df$y, frequency = 12, start = c(1949, 1))

# 4. Create a base forecast object and update it with Prophet's results
prophet_forecast_obj <- forecast(historical_ts)

# Replace base predictions with Prophet's values
prophet_forecast_obj$mean <- ts(
  prophet_forecast$yhat,
  start = start(historical_ts),
  frequency = frequency(historical_ts)
)

# Add Prophet's prediction intervals (adjust column names if needed)
prophet_forecast_obj$lower <- ts(
  cbind(`95%` = prophet_forecast$yhat_lower),
  start = start(historical_ts),
  frequency = frequency(historical_ts)
)

prophet_forecast_obj$upper <- ts(
  cbind(`95%` = prophet_forecast$yhat_upper),
  start = start(historical_ts),
  frequency = frequency(historical_ts)
)

# 5. Now use autoplot() as usual!
autoplot(prophet_forecast_obj)

Step 3: Optional - Forecast Only (No Historical Data) #

If you only want to plot the forecasted values (without historical data), you can construct a minimal forecast object:

# Extract only the forecast period values
forecast_period <- tail(prophet_forecast, 12) # Adjust 12 to match your forecast periods

# Create ts objects for predictions and intervals
forecast_ts <- ts(forecast_period$yhat, start = c(1961, 1), frequency = 12)
lower_ts <- ts(forecast_period$yhat_lower, start = c(1961, 1), frequency = 12)
upper_ts <- ts(forecast_period$yhat_upper, start = c(1961, 1), frequency = 12)

# Build the minimal forecast object
forecast_only_obj <- list(
  mean = forecast_ts,
  lower = cbind(`95%` = lower_ts),
  upper = cbind(`95%` = upper_ts),
  x = NULL, # No historical data
  series = "Prophet Forecast",
  model = NULL
)

# Assign the forecast class
class(forecast_only_obj) <- "forecast"

# Plot it!
autoplot(forecast_only_obj)

Key Notes #

• Always make sure the frequency and start parameters of your ts objects match your data's temporal frequency (monthly, daily, quarterly, etc.).
• Prophet's default prediction interval is 95%, but you can add 80% intervals too if you generated them when fitting the model.

内容的提问来源于stack exchange，提问作者Alex Kinman