使用 Arima 预测可能不起作用答案

【问题标题】：Using Arima forecast with possibly doesn't work使用 Arima 预测可能不起作用
【发布时间】：2020-03-27 16:37:46
【问题描述】：

我有两段代码：一个正在工作，另一个没有。也许有人知道为什么它不起作用。

PS。我知道这不是完全可重现的示例，但如果有必要，我会提供一个。

arima_cv = forecast_data_ini_split %>% 
  unnest(training_splits_for_cv_models) %>% 
  crossing(arima_models_for_cv %>% slice(17:18)) %>% 
  mutate(analysis_data = map(.x = splits, ~tk_analysis_fun(.x))) %>% 
  mutate(assessment_data = map(.x = splits, ~tk_assessment_fun(.x)))

这个有效

good = arima_cv %>% 
  mutate(models = furrr::future_pmap(list(analysis_data, p, d, q, P, D, Q),
                                     .f = ~Arima(y = ..1, order = c(..2, ..3, ..4), seasonal = c(..5, ..6, ..7), xreg = NULL,
                                                          include.mean = TRUE, include.drift = FALSE, method ="CSS-ML"))) %>% 
  mutate(models_metrics_ass = map2(.x = models, .y = assessment_data,  ~accuracy_assessment_fun(.x, .y)))

这个没有

bad = arima_cv %>% 
  mutate(models = pmap(list(analysis_data, p, d, q, P, D, Q),
                                     ~possibly(Arima, otherwise = NULL )(y = ..1, order = c(..2, ..3, ..4), seasonal = c(..5, ..6, ..7), xreg = NULL,
                                                 include.mean = TRUE, include.drift = FALSE, method ="ML"))) %>% 
  mutate(models_metrics_ass = map2(.x = models, .y = assessment_data,  ~accuracy_assessment_fun(.x, .y)))

错误信息是：

Error in eval(expr, p) : the ... list contains fewer than 4 elements

问题出在最后一行代码：

mutate(models_metrics_ass = map2(.x = models, .y = assessment_data,  ~accuracy_assessment_fun(.x, .y)))

似乎可能会以某种方式更改模型输出，而我无法做出预测和计数准确性。

提前致谢，缝制

【问题讨论】：

标签： r purrr arima forecast pmap

【解决方案1】：

所以，问题出在华宇身上，而不是可能。
我在 include.drift 参数中添加了一个常量，而它应该是一个变量。
顺便提一句。有可能和 pmap 的例子很少，所以如果你想要和工作的例子，请给我写信。

  unnest(training_splits_for_cv_models) %>%
  left_join(arima_models_for_cv) %>% 
  # crossing(arima_models_for_cv %>% 
  # slice(17:19) %>%
  mutate(analysis_data = map(.x = splits, ~tk_analysis_fun(.x))) %>% 
  mutate(assessment_data = map(.x = splits, ~tk_assessment_fun(.x))) %>% 
  mutate(models = furrr::future_pmap(list(analysis_data, p, d, q, P, D, Q, include.drift),
                                       .f = ~possibly(Arima, otherwise = NULL)(y = ..1, 
                                                   order = c(..2, ..3, ..4), 
                                                   seasonal = c(..5, ..6, ..7), 
                                                   include.drift = ..8, method ="CSS-ML"))) %>%
    mutate(models_metrics_ass = map2(.x = models, .y = assessment_data,  ~accuracy_assessment_fun(.x, .y)))

【讨论】：