R 尝试为包创建函数，当存在分类变量时，该函数会自动绘制给定模型的变量响应_R_Function_Ggplot2_Tidyverse

R 尝试为包创建函数，当存在分类变量时，该函数会自动绘制给定模型的变量响应

r function

R 尝试为包创建函数，当存在分类变量时，该函数会自动绘制给定模型的变量响应,r,function,ggplot2,tidyverse,R,Function,Ggplot2,Tidyverse,我试图创建一个函数，当至少一个变量是分类变量时，该函数将绘制所选变量的响应当所有变量都是数值变量时，我通常会保持所有其他变量的均值，然后改变目标变量，以下是mtcars的一个示例： library(tidyverse) data("mtcars") 首先，我将修改am变量，使其成为分类变量 mt2 <- mtcars %>% mutate(am = case_when(am == 0 ~ "Automatic", am == 1 ~ "Manual")) %>% sele

我试图创建一个函数，当至少一个变量是分类变量时，该函数将绘制所选变量的响应

当所有变量都是数值变量时，我通常会保持所有其他变量的均值，然后改变目标变量，以下是mtcars的一个示例：

library(tidyverse)

data("mtcars")

首先，我将修改am变量，使其成为分类变量

mt2 <- mtcars %>% mutate(am = case_when(am == 0 ~ "Automatic", am == 1 ~ "Manual")) %>% select(mpg, am, wt, hp)

这就是我遇到麻烦的时候当然，如果我在存在分类变量时尝试此操作，我会遇到问题，因为如果它尝试获取数据帧的分类值的平均值，则会失败：

model2 <- lm(mpg ~ wt + hp + am, data = mt2)

我会得到：

Error: variable 'am' was fitted with type "character" but type "numeric" was supplied

因此，我尝试了以下方法：

Plot_Response2 <- function(Model, variable){

  # First I get the names of all categorical variables
  Categoricals <- Model$model %>% mutate_if(is.factor, as.character) %>% select_if(is.character) %>% colnames()

  # generate a data.frame with all the means copied 20 times for each level

   Means <- Model$model %>% mutate_if(is.factor, as.character) %>% mutate_if(is.numeric, mean) %>% group_by_if(is.character) %>% summarise_if(is.numeric, mean) %>% ungroup()
   Means <- Means[rep(seq_len(nrow(Means)), each = 20),]  %>% arrange_if(is.character) %>% group_split(substitute(variable))

  return(Means)
}

我会得到：

Error: variable 'am' was fitted with type "character" but type "numeric" was supplied

我为此编写了以下代码，但无法将其添加到函数中：

Means <- model2$model %>% mutate_if(is.factor, as.character) %>% mutate_if(is.numeric, mean) %>% group_by_if(is.character) %>% summarise_if(is.numeric, mean) %>% ungroup()
Means <- Means[rep(seq_len(nrow(Means)), each = 20),]  %>% arrange_if(is.character) %>% group_split(am)

MinMax <- model2$model %>% select(wt) %>% pull(wt) %>% range()
MinMax <- seq(from = MinMax[1], to = MinMax[2], length.out = 20)

for(i in 1:length(Means)){
  Means[[i]]$wt <- MinMax
}

Means <- bind_rows(Means)
Means$Predicted <- predict(model2, Means)
Means$SE <- predict(model2, Means, se.fit = T)$se.fit

ggplot(Means, aes(x = wt, y = Predicted)) + geom_ribbon(aes(ymax = Predicted + SE, ymin = Predicted - SE, fill = am), alpha = 0.5) + geom_line(aes(color = am)) + theme_classic()

我会得到：

Error: variable 'am' was fitted with type "character" but type "numeric" was supplied

同样，我使用了这段代码，这段代码似乎无法与函数2结合在一起

Means <- model2$model %>% mutate_if(is.factor, as.character) %>% mutate_if(is.numeric, mean) %>% group_by_if(is.character) %>% summarise_if(is.numeric, mean) %>% ungroup()
Means <- Means[rep(seq_len(nrow(Means)), each = 20),]  %>% arrange_if(is.character) %>% group_split(am)

Means <- bind_rows(Means)
Means$Predicted <- predict(model2, Means)
Means$SE <- predict(model2, Means, se.fit = T)$se.fit

ggplot(Means, aes(x = am, y = Predicted)) + geom_errorbar(aes(ymin = Predicted - SE, ymax = Predicted + SE)) + geom_point() + theme_classic()

表示%mutate\u if（is.factor，as.character）%%>%mutate\u if（is.numeric，mean）%%>%groupby\u if（is.character）%%>%summary\u if（is.numeric，mean）%%>%ungroup（）
表示%arrange\u如果（是.字符）%>%group\u拆分（am）
意思是这里有一个版本，它使用了更多的tidyverse函数，只是为了让事情变得更简单
Plot_Response <- function(Model, variable, N=20) {
  model_data <- model.frame(Model)
  stopifnot(variable %in% names(model_data))

  # get all variables we need to dummy values for
  all_vars <- model_data %>% select(-one_of(variable))
  num_vars <- all_vars %>% select_if(is.numeric) %>% summarize_all(mean)
  cat_vars <- all_vars %>% select_if(Negate(is.numeric)) %>% purrr::map(unique)

  resp_var <- model_data %>% pull(variable) 
  if(is.numeric(resp_var)) {
    resp_vals <- seq(min(resp_var), max(resp_var), length.out=N)
  } else {
    resp_vals <- unique(resp_var)
  }

  new_data <- tidyr::crossing(num_vars, !!!cat_vars, !!variable:=resp_vals)

  pred <- broom::augment(Model, newdata = new_data, se_fit=TRUE)

  ## Plot the response
  my_aes <- aes(x= !!sym(variable), y = .fitted)
  if (length(cat_vars)==1) {
    my_aes[["fill"]] <- sym(names(cat_vars))
  } else if (length(cat_vars)>1) {
    my_aes[["fill"]] <- quo(interaction(!!!syms(names(cat_vars))))
  }
  range_aes <- aes(ymax= .fitted + .se.fit, ymin = .fitted - .se.fit)
  result <- ggplot(pred, my_aes) + theme_classic() + ylab("Predicted")
  if(is.numeric(resp_var)) {
    result + 
      (if (length(cat_vars)>0) {
        geom_ribbon(range_aes) 
      } else {
        geom_ribbon(range_aes, fill="grey")
      }) + 
      geom_line()
  } else {
    result + 
      geom_errorbar(range_aes) + 
      geom_point() 
  } 

}

Plot\u响应%purrr:：map（唯一）
响应变量%pull（变量）
如果（是数字（对应变量））{
相应的我相信你所描述的正是该软件包解决的那种问题。软件包描述的第一行很好地说明了这一点：
DescTools是各种基本统计数据的广泛集合
R基本系统中不提供的功能和舒适性包装
数据的有效描述
我不喜欢加载一串包来完成R中的工作。但是，我对这一个例外。我认为Andri Signorell的大量工具集确实非常出色。在DescTools
中定义的函数与tidyverse
中定义的函数之间可能存在冲突，因此我用你的话来概括我的答案t求助于tidyverse

# DescTools needs to be available
  if (!require(DescTools)) {
    install.packages("DescTools")
  }
  library(DescTools)

# Create factors in mtcars
  mt3 <- mtcars
  mt3$am <- factor(mt3$am, labels = c("man", "auto"))
  mt3$vs <- factor(mt3$vs, labels = c("v", "str"))




还有两个例子说明了对于相互作用的因素，甚至对于单个连续变量，这是多么简单
  Desc(mpg ~ am:vs, mt3)
  Desc(mt3$qsec)  


也许结帐已经做了很多。谢谢@MrFlick，我甚至还学习了一门我不知道的交叉。你有我所需要的一切
Means <- model2$model %>% mutate_if(is.factor, as.character) %>% mutate_if(is.numeric, mean) %>% group_by_if(is.character) %>% summarise_if(is.numeric, mean) %>% ungroup()
Means <- Means[rep(seq_len(nrow(Means)), each = 20),]  %>% arrange_if(is.character) %>% group_split(am)

Means <- bind_rows(Means)
Means$Predicted <- predict(model2, Means)
Means$SE <- predict(model2, Means, se.fit = T)$se.fit

ggplot(Means, aes(x = am, y = Predicted)) + geom_errorbar(aes(ymin = Predicted - SE, ymax = Predicted + SE)) + geom_point() + theme_classic()

Plot_Response <- function(Model, variable, N=20) {
  model_data <- model.frame(Model)
  stopifnot(variable %in% names(model_data))

  # get all variables we need to dummy values for
  all_vars <- model_data %>% select(-one_of(variable))
  num_vars <- all_vars %>% select_if(is.numeric) %>% summarize_all(mean)
  cat_vars <- all_vars %>% select_if(Negate(is.numeric)) %>% purrr::map(unique)

  resp_var <- model_data %>% pull(variable) 
  if(is.numeric(resp_var)) {
    resp_vals <- seq(min(resp_var), max(resp_var), length.out=N)
  } else {
    resp_vals <- unique(resp_var)
  }

  new_data <- tidyr::crossing(num_vars, !!!cat_vars, !!variable:=resp_vals)

  pred <- broom::augment(Model, newdata = new_data, se_fit=TRUE)

  ## Plot the response
  my_aes <- aes(x= !!sym(variable), y = .fitted)
  if (length(cat_vars)==1) {
    my_aes[["fill"]] <- sym(names(cat_vars))
  } else if (length(cat_vars)>1) {
    my_aes[["fill"]] <- quo(interaction(!!!syms(names(cat_vars))))
  }
  range_aes <- aes(ymax= .fitted + .se.fit, ymin = .fitted - .se.fit)
  result <- ggplot(pred, my_aes) + theme_classic() + ylab("Predicted")
  if(is.numeric(resp_var)) {
    result + 
      (if (length(cat_vars)>0) {
        geom_ribbon(range_aes) 
      } else {
        geom_ribbon(range_aes, fill="grey")
      }) + 
      geom_line()
  } else {
    result + 
      geom_errorbar(range_aes) + 
      geom_point() 
  } 

}

model1 <- lm(mpg ~ wt + hp + am, data = mt2)
Plot_Response(model1, "wt")
Plot_Response(model1, "am")

# DescTools needs to be available
  if (!require(DescTools)) {
    install.packages("DescTools")
  }
  library(DescTools)

# Create factors in mtcars
  mt3 <- mtcars
  mt3$am <- factor(mt3$am, labels = c("man", "auto"))
  mt3$vs <- factor(mt3$vs, labels = c("v", "str"))

# mpg as a function of weight
  dev.new(width = 6, height = 4.5)
  opar <- par(mfrow = c(1, 2))
  Desc(mpg ~ wt, mt3, main = "Manual", subset = am == "man")
  Desc(mpg ~ wt, mt3, main = "Automatic", subset = am == "auto")
  par(opar)

# mpg as a function of transmission
  Desc(mpg ~ am, mt3)

  Desc(mpg ~ am:vs, mt3)
  Desc(mt3$qsec)