GLM适合R之前的组织因子变量

问题描述 投票：1回答：1

我在R上的GLM上做了很多工作，涉及的是相当大的数据集（通常在模型拟合中包含数十个变量）。为了能够在拟合模型后产生某种类型的图形输出，我发现“准备”我要作为因子拟合的任何变量（那些名称以下面的f_开头）非常有用。在模型拟合之前。我的意思是：

（（i）因为我在拟合GLM之前对每个因子进行了重新排序，以使参考水平等于权重最大的水平，所以我想在relevel()命令之前保留水平顺序；

（（ii）为了稍后在图形中突出显示参考水平，我希望将其记录保存在单独的变量中。

我使用内置的mtcars数据集将这种方法的示例放在一起。

当前我有此代码：

library(dplyr)

data(mtcars)

# tidy up and make easier to read

df <- mtcars # built in data set

# let's make it a bit easier to follow

df <- df %>%
  select(mpg,
         f_cylinders = cyl,
         c_displacement = disp,
         c_hp = hp,
         c_final_drive_ratio = drat,
         c_weight = wt,
         c_qtr_mile_time = qsec,
         f_v_or_straight = vs,
         f_transmission = am,
         f_gears = gear,
         f_num_carbs = carb)

df$f_v_or_straight <- ifelse(df$f_v_or_straight == 0, "V", "Straight")
df$f_transmission <- ifelse(df$f_transmission == 0, "Automatic", "Manual")

df$glm_weight <- 1

# organise factors - levels, reference level, weights

my_list <- list()

df$f_cylinders <- as.factor(df$f_cylinders)
my_list$f_cylinders_levels <- levels(df$f_cylinders) 
my_list$f_cylinders_weights <- df %>% group_by(f_cylinders) %>% summarise(glm_weight = sum(glm_weight)) %>% ungroup() %>% pull(glm_weight)
my_list$f_cylinders_ref <- "8"
df$f_cylinders <- df$f_cylinders %>% relevel(ref = my_list$f_cylinders_ref)

df$f_v_or_straight <- as.factor(df$f_v_or_straight)
my_list$f_v_or_straight_levels <- levels(df$f_v_or_straight) 
my_list$f_v_or_straight_weights <- df %>% group_by(f_v_or_straight) %>% summarise(glm_weight = sum(glm_weight)) %>% ungroup() %>% pull(glm_weight)
my_list$f_v_or_straight_ref <- "V"
df$f_v_or_straight <- df$f_v_or_straight %>% relevel(ref = my_list$f_v_or_straight_ref)

df$f_transmission <- as.factor(df$f_transmission)
my_list$f_transmission_levels <- levels(df$f_transmission) 
my_list$f_transmission_weights <- df %>% group_by(f_transmission) %>% summarise(glm_weight = sum(glm_weight)) %>% ungroup() %>% pull(glm_weight)
my_list$f_transmission_ref <- "Automatic"
df$f_transmission <- df$f_transmission %>% relevel(ref = my_list$f_transmission_ref)

df$f_gears <- as.factor(df$f_gears)
my_list$f_gears_levels <- levels(df$f_gears) 
my_list$f_gears_weights <- df %>% group_by(f_gears) %>% summarise(glm_weight = sum(glm_weight)) %>% ungroup() %>% pull(glm_weight)
my_list$f_gears_ref <- "3"
df$f_gears <- df$f_gears %>% relevel(ref = my_list$f_gears_ref)

df$f_num_carbs <- as.factor(df$f_num_carbs)
my_list$f_num_carbs_levels <- levels(df$f_num_carbs) 
my_list$f_num_carbs_weights <- df %>% group_by(f_num_carbs) %>% summarise(glm_weight = sum(glm_weight)) %>% ungroup() %>% pull(glm_weight)
my_list$f_num_carbs_ref <- "4"
df$f_num_carbs <- df$f_num_carbs %>% relevel(ref = my_list$f_num_carbs_ref)

此代码工作正常，但是...在现实世界中，我正在处理数十个因子变量，而不仅仅是上面的5个。因此，如果我有50个因子变量，那么我做同样的事情超过50次。我想将此准备捆绑到一个函数调用中，本质上说：

对于每个名称以f_开头（即看起来像f_xxx）的字段：

  1. 将其从chr /int/转换为因子f_xxx;

  2. 锻炼体重f_xxx_weights

  3. 制定参考水平f_xxx_ref（不确定牵头关系时该怎么做）；

  4. 将当前因子水平存储在f_xxx_levels中；

  5. 重新排列因子级别，以使f_xxx_ref在列表中排在首位。

我在这里要问的很多...但是最感激我前进的任何事情，将不胜感激。

谢谢。

r function
1个回答
0
投票

考虑使用用户定义的方法对relevel进程进行泛化，然后使用f_映射通过purrr::map_df列调用您的进程（与整洁一致）：

数据

library(dplyr)
library(purrr)

df <- mtcars %>%
  select(mpg,
         f_cylinders = cyl,
         c_displacement = disp,
         c_hp = hp,
         c_final_drive_ratio = drat,
         c_weight = wt,
         c_qtr_mile_time = qsec,
         f_v_or_straight = vs,
         f_transmission = am,
         f_gears = gear,
         f_num_carbs = carb) %>%
  mutate(f_v_or_straight = ifelse(f_v_or_straight == 0, 
                                  "V", 
                                  "Straight"),
         f_transmission = ifelse(f_transmission == 0, 
                                 "Automatic", 
                                 "Manual"),
         glm_weight = 1)

rlevel处理 （（使用table频率）

proc_rlevel <- function(col) {
    agg <- df %>% group_by_at(col) %>% 
      summarise(glm_weight = sum(glm_weight)) %>%
      arrange(desc(glm_weight))

    f_ref <- df[[col]] %>% 
      as.character() %>%
      as.factor() %>%
      relevel(ref = paste(agg[[col]][1]))

    return(f_ref)
}

# REPLACING ORIGINAL f_cols WITH TWO WAY PIPES
df[grep("f_", names(df))] %<>% 
  names() %>%
  setNames(identity(.)) %>%
  map_df(proc_rlevel)

查看更改

# ORIGINAL LEVELS
df %>% 
  select(starts_with("f_")) %>%
  map_df(as.factor) %>%
  map(levels)
# $f_cylinders
# [1] "4" "6" "8"
# 
# $f_v_or_straight
# [1] "Straight" "V"       
# 
# $f_transmission
# [1] "Automatic" "Manual"   
# 
# $f_gears
# [1] "3" "4" "5"
# 
# $f_num_carbs
# [1] "1" "2" "3" "4" "6" "8"


# ADJUSTED LEVELS
df %>% 
  select(starts_with("f_")) %>%
  map_df(as.factor) %>%
  map(levels)
# $f_cylinders
# [1] "8" "4" "6"
# 
# $f_v_or_straight
# [1] "V"        "Straight"
# 
# $f_transmission
# [1] "Automatic" "Manual"   
# 
# $f_gears
# [1] "3" "4" "5"
# 
# $f_num_carbs
# [1] "2" "1" "3" "4" "6" "8"
最新问题
© www.soinside.com 2019 - 2024. All rights reserved.