如何在 R 中构建和测试多个模型

Question

假设我有一个数据集（忘记分布）：

modData <- data.frame("A" = rnorm(20, 15, 3),
                      "B" = rnorm(20, 20, 3),
                      "C" = rnorm(20, 25, 3),
                      "X" = rnorm(20, 5, 1)
                      )

如果我分别X用作预测变量A、B和C响应：

md1 <- lm(A ~ X, data = modData)
md2 <- lm(B ~ X, data = modData)
md3 <- lm(C ~ X, data = modData)

然后对每个模型进行 Shapiro 测试和 boxcox 测试，例如：

shapiro.test(residuals(md1))
boxcox(md1, plotit = T)

有没有一种方便的方法来构建和测试多个模型而无需手动输入每个模型？

Answer 1

这是使用的替代方法tidyverse：

modData <- data.frame("A" = rnorm(20, 15, 3),
                      "B" = rnorm(20, 20, 3),
                      "C" = rnorm(20, 25, 3),
                      "X" = rnorm(20, 5, 1))
library(tidyverse)
library(broom)

# specify predictor and target variables
x = "X"
y = names(modData)[names(modData)!= x]

expand.grid(y,x) %>%                                    # create combinations
  mutate(model_id = row_number(),                       # create model id
         frml = paste0(Var1, "~", Var2)) %>%            # create model formula
  group_by(model_id, Var1, Var2) %>%                    # group by the above
  nest() %>%                                            # nest data
  mutate(m = map(data, ~lm(.$frml, data = modData)),    # create models
         m_table = map(m, ~tidy(.)),                    # tidy model output
         st = map(m, ~shapiro.test(residuals(.)))) -> dt_model_info  # shapiro test

# access model info
dt_model_info
dt_model_info$m
dt_model_info$m_table
dt_model_info$st

# another way to access info
dt_model_info %>% unnest(m_table)