使用 List-Columns 和 Purrr 拟合许多模型时出错答案

【问题标题】：Error while fitting many models using List-Columns and Purrr使用 List-Columns 和 Purrr 拟合许多模型时出错
【发布时间】：2017-07-05 11:15:59
【问题描述】：

我正在尝试执行来自：http://unconj.ca/blog/forecasting-yyz-passengers-in-the-tidyverse.html的代码

在中间的某个地方.....

首先，我们创建模型及其参数的数据集。以下将生成所有可能的 ETS 模型的数据框

ets.params <- crossing(error = c("A", "M"), trend = c("N", "A", "M"),seasonal = c("N", "A", "M"), damped = c(TRUE, FALSE)) %>%
    # Drop combinations with a damped non-trend.
    mutate(drop = ifelse(trend == "N" & damped, TRUE, FALSE)) %>% 
    # Create labels for the models out of these parameters.
    mutate(kind = "ETS", desc = paste0("(", error, ",", trend,ifelse(damped, "d", ""),",", seasonal, ")"),model = paste0(error, trend, seasonal)) 
    # Drop nonsensical models (these are flagged by `ets` anyway).
ets.params1<-subset(ets.params,!(ets.params$model %in% c("MMA","AMN","AMA","AMM","ANM","AAM")))

ets.params1

# A tibble: 24 x 8
   error trend seasonal damped  drop  kind     desc model
   <chr> <chr>    <chr>  <lgl> <lgl> <chr>    <chr> <chr>
 1     A     A        A  FALSE FALSE   ETS  (A,A,A)   AAA
 2     A     A        A   TRUE FALSE   ETS (A,Ad,A)   AAA
 3     A     A        N  FALSE FALSE   ETS  (A,A,N)   AAN
 4     A     A        N   TRUE FALSE   ETS (A,Ad,N)   AAN
 5     A     N        A  FALSE FALSE   ETS  (A,N,A)   ANA
 6     A     N        A   TRUE  TRUE   ETS (A,Nd,A)   ANA
 7     A     N        N  FALSE FALSE   ETS  (A,N,N)   ANN
 8     A     N        N   TRUE  TRUE   ETS (A,Nd,N)   ANN
 9     M     A        A  FALSE FALSE   ETS  (M,A,A)   MAA
10     M     A        A   TRUE FALSE   ETS (M,Ad,A)   MAA
# ... with 14 more rows

有了这组模型参数，我们可以创建列表列，其中包含训练数据和计算模型的函数（在本例中为 ets）。

我可以执行上面的代码，但是下面的代码会报错...

ets.models <- ets.params %>%
    # Add in the training set and the modelling function.
    mutate(fn = replicate(forecast::ets, n = n()),
           train = replicate(list(train), n = n())) %>%
    # Create a "param" column to pass to `fn`.
    mutate(params = purrr::transpose(list(
        "y" = train, "model" = model, "damped" = damped
    ))) %>%
    select(kind, desc, train, fn, params)

ets.models

我收到错误：错误：不应直接调用此函数。错误似乎来自 mutate()..

需要帮助！！！！

【问题讨论】：

标签： r time-series tidyverse purrr

【解决方案1】：

看来您遇到了一个有趣的问题。如果您能提供一个可重现的示例会很好，但这是我获取源数据并将其处理到ets 模型列表中的看法

library(tidyverse)
library(readxl)

if (!file.exists("bulletin_opendata.xlsx"))
  download.file("http://opendata.toronto.ca/it/com/bulletin_opendata.xlsx", 
                "bulletin_opendata.xlsx")

y <- read_excel("bulletin_opendata.xlsx", sheet = "Trans", 
              col_names = FALSE, skip=256) %>% # read xlsx file
   mutate(X7=X7*1e3) %>% pull(X7) %>% ts(f=12) # extract column as ts

library(forecast)

ets.params <- crossing(error = c("A", "M"), trend = c("N", "A", "M"),
                       seasonal = c("N", "A", "M"), damped = c(TRUE, FALSE)) %>%
  # Drop combinations with a damped non-trend.
  mutate(drop = ifelse(trend == "N" & damped, TRUE, FALSE)) %>% 
  # Create labels for the models out of these parameters.
  mutate(kind = "ETS", 
         desc = paste0("(", error, ",", trend, ifelse(damped, "d", ""),",", seasonal, ")"), 
         model = paste0(error, trend, seasonal)) 
# Drop nonsensical models (these are flagged by `ets` anyway).
ets.params1<-subset(ets.params,!(ets.params$model %in% c("MMA","AMN","AMA","AMM","ANM","AAM")))

illegal.models<-c("(A,Nd,A)", "(A,Nd,N)", "(M,Nd,A)", "(M,Nd,M)", "(M,Nd,N)")

ets.params1 %>% mutate(y=list(y)) %>%     # add train data
  filter(!desc %in% illegal.models) %>%   # some of the model combinations are not possible with ETS
  select(y, model, damped) %>%            # prepare a list to be mapped to ets
  mutate(mod = pmap(.,ets),               # map a model
         prog = map(mod, forecast, h=12)) # and forecast 1 year ahead.

【讨论】：