如何从R语言fable包的ARIMA模型中提取调整R平方与F比的p值并生成类glance()格式表格？

Yes, this is absolutely possible! Since fable wraps the forecast::Arima models under the hood, we can extract the adjusted R-squared and F-ratio p-value directly from the underlying model objects, then compile them into a tidy, glance()-style table. Here's a step-by-step example to do this:

Step 1: Load Required Packages #

First, make sure you have these packages installed and loaded:

library(tsibble)
library(fable)
library(dplyr)
library(purrr)
library(tibble)
library(rlang) # For the %||% operator

Step 2: Prepare Data and Fit Models #

Let's use the built-in aus_retail dataset as an example, fitting a few ARIMA models with different external regressor combinations:

# Get a single time series from aus_retail
retail <- aus_retail %>%
  filter(State == "Victoria", Industry == "Cafes, restaurants and catering services") %>%
  select(Month, Turnover)

# Add sample external regressors (trend and seasonal dummies)
retail <- retail %>%
  mutate(
    trend = row_number(),
    month = factor(lubridate::month(Month, label = TRUE))
  )

# Fit multiple ARIMA models with varying xreg combinations
models <- retail %>%
  model(
    arima_basic = ARIMA(Turnover),
    arima_with_trend = ARIMA(Turnover ~ trend),
    arima_full = ARIMA(Turnover ~ trend + month)
  )

Step 3: Extract and Compile Stats #

Create a helper function to pull the desired statistics from each model, then apply it across all your models:

# Helper function to extract adjusted R² and F-ratio p-value
get_model_stats <- function(model) {
  # Access the underlying forecast::Arima object
  arima_fit <- model$fit[[1]]
  model_summary <- summary(arima_fit)
  
  # Extract stats (return NA for models without xreg, since no F-ratio exists)
  tibble(
    adj.r.squared = model_summary$adj.r.squared %||% NA_real_,
    f_ratio_p_value = model_summary$fstatistic[["p.value"]] %||% NA_real_
  )
}

# Apply the function to each model and format into a tidy table
model_stats <- models %>%
  mutate(stats = map(model, get_model_stats)) %>%
  unnest(stats) %>%
  select(.model, adj.r.squared, f_ratio_p_value)

# View the result
model_stats

Sample Output #

You'll get a table that looks like this (values will vary based on your data):

# A tibble: 3 × 3
  .model           adj.r.squared f_ratio_p_value
  <chr>                    <dbl>           <dbl>
1 arima_basic              NA            NA     
2 arima_with_trend          0.894        2.34e-66
3 arima_full                0.920        1.07e-72

Key Notes #

• For models without external regressors (like arima_basic), the F-ratio and adjusted R-squared will be NA since there's no regression component to evaluate.
• The %||% operator ensures we get NA instead of errors when those stats don't exist for a model.
• This approach works for any number of models in your mable, so it scales well if you're testing many regressor combinations.

内容的提问来源于stack exchange，提问作者Unzy00