dplyr：评估错误：对象“。”用 gamlss 找不到，但用 lm、gam、glm 方法都很好答案

【问题标题】：dplyr: Evaluation error: object '.' not found with gamlss but all good with lm, gam, glm methodsdplyr：评估错误：对象“。”用 gamlss 找不到，但用 lm、gam、glm 方法都很好
【发布时间】：2018-02-25 22:48:23
【问题描述】：

上下文：tidyverse 和 dplyr 环境/工作流。

如果我在尝试处理回归结果集合时遇到了以下问题，我将不胜感激。

这个最小的可重复性显示了问题

mtcars %>% 
  gamlss(mpg ~ hp + wt + disp, data = .) %>% 
  model.frame()

下面的示例说明了更广泛的背景并按预期工作（生成显示的图像）。如果我所做的只是将~lm(...) 更改为~glm(...) 或~gam(...)，它也可以工作：

library(tidyverse)
library(broom)
library(gamlss)

library(datasets)

mtcars %>% 
  nest(-am) %>% 
  mutate(am = factor(am, levels = c(0, 1), labels = c("automatic", "manual")),
         fit = map(data, ~lm(mpg ~ hp + wt + disp, data = .)),
         results = map(fit, augment)) %>% 
  unnest(results) %>% 
  ggplot(aes(x = mpg, y = .fitted)) +
    geom_abline(intercept = 0, slope = 1, alpha = .2) +  # Line of perfect fit
    geom_point() +
    facet_grid(am ~ .) +
    labs(x = "Miles Per Gallon", y = "Predicted Value") +
    theme_bw()

但是，如果我尝试如下使用~gamlss(...)：

mtcars %>% 
  nest(-am) %>% 
  mutate(am = factor(am, levels = c(0, 1), labels = c("automatic", "manual")),
         fit = map(data, ~gamlss(mpg ~ hp + wt + disp, data = .)),
         results = map(fit, augment)) %>% 
  unnest(results) %>% 
  ggplot(aes(x = mpg, y = .fitted)) +
    geom_abline(intercept = 0, slope = 1, alpha = .2) +  # Line of perfect fit
    geom_point() +
    facet_grid(am ~ .) +
    labs(x = "Miles Per Gallon", y = "Predicted Value") +
    theme_bw()

我观察到以下错误：

GAMLSS-RS iteration 1: Global Deviance = 58.7658 
GAMLSS-RS iteration 2: Global Deviance = 58.7658 
GAMLSS-RS iteration 1: Global Deviance = 76.2281 
GAMLSS-RS iteration 2: Global Deviance = 76.2281 
******************************************************************
Family:  c("NO", "Normal") 

Call:  gamlss(formula = mpg ~ hp + wt + disp, data = .) 

Fitting method: RS() 

------------------------------------------------------------------
Mu link function:  identity
Mu Coefficients:
             Estimate Std. Error t value Pr(>|t|)    
(Intercept) 43.811721   3.387118  12.935 4.05e-07 ***
hp           0.001768   0.021357   0.083  0.93584    
wt          -6.982534   1.998827  -3.493  0.00679 ** 
disp        -0.019569   0.021460  -0.912  0.38559    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

------------------------------------------------------------------
Sigma link function:  log
Sigma Coefficients:
            Estimate Std. Error t value Pr(>|t|)   
(Intercept)   0.8413     0.1961    4.29  0.00105 **
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

------------------------------------------------------------------
No. of observations in the fit:  13 
Degrees of Freedom for the fit:  5
      Residual Deg. of Freedom:  8 
                      at cycle:  2 

Global Deviance:     58.76579 
            AIC:     68.76579 
            SBC:     71.59054 
******************************************************************
Error in mutate_impl(.data, dots) : 
  Evaluation error: object '.' not found.
In addition: Warning messages:
1: Deprecated: please use `purrr::possibly()` instead 
2: Deprecated: please use `purrr::possibly()` instead 
3: Deprecated: please use `purrr::possibly()` instead 
4: Deprecated: please use `purrr::possibly()` instead 
5: Deprecated: please use `purrr::possibly()` instead 
6: In summary.gamlss(model) :
  summary: vcov has failed, option qr is used instead

15: stop(list(message = "Evaluation error: object '.' not found.", 
        call = mutate_impl(.data, dots), cppstack = NULL))
14: .Call(`_dplyr_mutate_impl`, df, dots)
13: mutate_impl(.data, dots)
12: mutate.tbl_df(tbl_df(.data), ...)
11: mutate(tbl_df(.data), ...)
10: as.data.frame(mutate(tbl_df(.data), ...))
9: mutate.data.frame(., am = factor(am, levels = c(0, 1), labels = c("automatic", 
       "manual")), fit = map(data, ~gamlss(mpg ~ hp + wt + disp, 
       data = .)), results = map(fit, augment))
8: mutate(., am = factor(am, levels = c(0, 1), labels = c("automatic", 
       "manual")), fit = map(data, ~gamlss(mpg ~ hp + wt + disp, 
       data = .)), results = map(fit, augment))
7: function_list[[i]](value)
6: freduce(value, `_function_list`)
5: `_fseq`(`_lhs`)
4: eval(quote(`_fseq`(`_lhs`)), env, env)
3: eval(quote(`_fseq`(`_lhs`)), env, env)
2: withVisible(eval(quote(`_fseq`(`_lhs`)), env, env))
1: mtcars %>% nest(-am) %>% mutate(am = factor(am, levels = c(0, 
       1), labels = c("automatic", "manual")), fit = map(data, ~gamlss(mpg ~ 
       hp + wt + disp, data = .)), results = map(fit, augment)) %>% 
       unnest(results) %>% ggplot(aes(x = mpg, y = .fitted))

是否有人建议需要更改哪些内容才能使此示例按预期工作？

如果您能深入了解问题所在，我将不胜感激。为什么它不起作用。如何诊断此类问题。

【问题讨论】：

您确定您的示例完全可重现吗？运行第一部分时出现错误Error in mutate_impl(.data, dots) : Evaluation error: object 'augment' not found.。我可以确认我安装了最新版本的 tidyverse。
道歉。我以为 tidyverse 加载了扫帚。我在示例中添加了 library(broom)。
idk，但这不是特别map 和gamlss 一起出现的问题。要测试它，请尝试 mtcars %>% split(mtcars$am) %>% map(~ gamlss(mpg ~ hp + wt + disp, data = .)) 给出 gamlss 没有错误的结果。
重现错误的更快方法：mtcars %>% gamlss(mpg ~ hp + wt + disp, data = .) %>% model.frame()。我不认为model.frame.gamlss 实现与broom 使用的其他model.frame 函数一致。但这只是猜测。
是的，我认为 gamlss 不能与 broom 一起使用。对不起????‍♂️

标签： r dplyr gamlss

【解决方案1】：

model.frame.gamlss 没有正确考虑 data 参数的原始环境。

在下面的代码中查看我的 cmets：

model.frame.gamlss <- function(formula, what = c("mu", "sigma", "nu", "tau"), parameter = NULL, ...) 
{
    object <- formula
    dots <- list(...)
    what <- if (!is.null(parameter)) {
        match.arg(parameter, choices = c("mu", "sigma", "nu", "tau"))
    } else match.arg(what)
    Call <- object$call
    parform <- formula(object, what)
    data <- if (!is.null(Call$data)) {
        # problem here, as Call$data is .
        eval(Call$data)
        # instead, this would work:
        # eval(Call$data, environment(formula$mu.terms))
        # (there is no formula$terms, just mu.terms and sigma.terms)
    } else {
        environment(formula$terms)
    }
    Terms <- terms(parform)
    mf <- model.frame(
        Terms, 
        data, 
        xlev = object[[paste(what, "xlevels", sep = ".")]]
    )
    mf
}

我想应该向gamlss 维护者提交问题，除非已经这样做了。

【讨论】：

你能分享识别那行代码的调试工具吗？我遇到了一个后续错误，我也会将其作为另一个问题，您可能也知道答案。
在这种情况下，我真的只是在看代码......但也许trace 是一个可以帮助你的功能，例如stackoverflow.com/questions/48757383/…

【解决方案2】：

基于 RolnadASc 的部分答案...

原始问题的中间数据集和图表输出由以下再现。使用这种方法的一个改进是我们不需要创建fit 的中间步骤。

library(tidyverse)
library(broom)
library(gamlss)
library(datasets)

model.frame.gamlss <- function(formula, what = c("mu", "sigma", "nu", "tau"), parameter = NULL, ...) 
{
    object <- formula
    dots <- list(...)
    what <- if (!is.null(parameter)) {
        match.arg(parameter, choices = c("mu", "sigma", "nu", "tau"))
    } else match.arg(what)
    Call <- object$call
    parform <- formula(object, what)
    data <- if (!is.null(Call$data)) {
        # problem here, as Call$data is .
        # eval(Call$data)
        # instead, this would work:
        eval(Call$data, environment(formula$mu.terms))
        # (there is no formula$terms, just mu.terms and sigma.terms)
    } else {
        environment(formula$terms)
    }
    Terms <- terms(parform)
    mf <- model.frame(
        Terms, 
        data, 
        xlev = object[[paste(what, "xlevels", sep = ".")]]
    )
    mf
}
aug_func <- function(df){
          augment(gamlss(mpg ~ hp + wt + disp, data=df))
        }

mtcars %>% 
  mutate(am = factor(am, levels = c(0, 1), labels = c("automatic", "manual"))) %>%
  group_by(am) %>%
  do(aug=aug_func(df=.)) %>%
  unnest(aug) %>%
    ggplot(aes(x = mpg, y = .fitted)) +
      geom_abline(intercept = 0, slope = 1, alpha = .2) +  # Line of perfect fit
      geom_point() +
      facet_grid(am ~ .) +
      labs(x = "Miles Per Gallon gamlss", y = "Predicted Value gamlss") +
      theme_bw()

【讨论】：