从带系数 (R) 的 RIDGE、LASSO 和净弹性回归中为许多变量提取公式

Question

我正在尝试修改我在这个 post 的一个答案中找到的一些代码：

Extract Formula From lm with Coefficients (R)

AlexB 提供了这些精彩的代码行：

get_formula <- function(model) {
  broom::tidy(model)[, 1:2] %>%
    mutate(sign = ifelse(sign(estimate) == 1, ' + ', ' - ')) %>% #coeff signs
    mutate_if(is.numeric, ~ abs(round(., 2))) %>% #for improving formatting
    mutate(a = ifelse(term == '(Intercept)', paste0('y ~ ', estimate), paste0(sign, estimate, ' * ', term))) %>%
    summarise(formula = paste(a, collapse = '')) %>%
    as.character
}

虽然这适用于我的一些代码，但我在使用 RIDGE、LASSO 和 Net Elastic Regression 调整它以从 glmnet 模型打印公式时遇到问题。

下面附上了我要提供的示例：

library(caret)
library(glmnet)
library(mlbench)
library(psych)
data("BostonHousing")
data <- BostonHousing
set.seed(23)
ind <- sample(2, nrow(data), replace = T, prob = c(0.7, 0.3))
train <- data[ind==1,]
test <- data[ind==2,]
custom <- trainControl(method = "repeatedcv",number = 10,repeats = 5,verboseIter = T)
set.seed(23)
ridge <- train(medv~., train,method = "glmnet",tuneGrid = expand.grid(alpha = 0,lambda = seq(0.0001,1,length = 5)),trControl = custom)
ridge
coef(ridge$finalModel, ridge$bestTune$lambda) # the coefficient estimates

get_formula <- function(model) {
  broom::tidy(model)[, 1:2] %>%
    mutate(sign = ifelse(sign(estimate) == 1, ' + ', ' - ')) %>% #coeff signs
    mutate_if(is.numeric, ~ abs(round(., 2))) %>% #for improving formatting
    mutate(a = ifelse(term == '(Intercept)', paste0('y ~ ', estimate), paste0(sign, estimate, ' * ', term))) %>%
    summarise(formula = paste(a, collapse = '')) %>%
    as.character
}
get_formula(ridge$finalModel)

但是，鉴于它与之前的格式不同 post，我在修改函数时遇到问题，以便它可以打印出我正在寻找的方程式。

给出错误：

Error: Problem with `mutate()` input `sign`.
x object 'estimate' not found
i Input `sign` is `ifelse(sign(estimate) == 1, " + ", " - ")`.
Run `rlang::last_error()` to see where the error occurred.

感谢您的帮助。

Answer 1

broom 包有 a tidy variant for glmnet - 您不需要使用 [, 1:2].

索引整理数据

只需使用 tidy(model) 管道的其余部分就可以正常工作。

这里是函数的关键部分，拿出来演示一下：


broom::tidy(ridge$finalModel) %>%
  mutate(sign = ifelse(sign(estimate) == 1, ' + ', ' - ')) %>% #coeff signs
  mutate_if(is.numeric, ~ abs(round(., 2))) %>% #for improving formatting
  mutate(a = ifelse(term == '(Intercept)', paste0('y ~ ', estimate), paste0(sign, estimate, ' * ', term))) 

# A tibble: 1,400 x 7
   term         step estimate lambda dev.ratio sign  a        
   <chr>       <dbl>    <dbl>  <dbl>     <dbl> <chr> <chr>    
 1 (Intercept)     1     21.7  6655.      0    " + " y ~ 21.68
 2 (Intercept)     2     21.7  6064.      0.01 " + " y ~ 21.73
 3 (Intercept)     3     21.7  5525.      0.01 " + " y ~ 21.73
 4 (Intercept)     4     21.7  5034.      0.01 " + " y ~ 21.74
 5 (Intercept)     5     21.7  4587.      0.01 " + " y ~ 21.74
 6 (Intercept)     6     21.8  4180.      0.01 " + " y ~ 21.75
 7 (Intercept)     7     21.8  3808.      0.01 " + " y ~ 21.75
 8 (Intercept)     8     21.8  3470.      0.01 " + " y ~ 21.76
 9 (Intercept)     9     21.8  3162.      0.01 " + " y ~ 21.77
10 (Intercept)    10     21.8  2881.      0.02 " + " y ~ 21.78
# … with 1,390 more rows

小提示：across 现在可以替换 mutate_if，例如

mutate(across(where(is.numeric), ~abs(round(., 2))))

Answer 2

稍作更新，即可得到岭回归方程，如下：

as.matrix(coef(ridge$finalModel, ridge$bestTune$lambda)) %>%
  as.data.frame() %>%
  tibble::rownames_to_column('term') %>%
  rename(estimate = 2) %>%
  mutate(sign = ifelse(sign(estimate) == 1, ' + ', ' - ')) %>% #coeff signs
  mutate(across(where(is.numeric), ~abs(round(., 2)))) %>% #for improving formatting
  mutate(a = ifelse(term == '(Intercept)', paste0('y ~ ', estimate), paste0(sign, estimate, ' * ', term))) %>%
  summarise(formula = paste(a, collapse = ''))

从带系数 (R) 的 RIDGE、LASSO 和净弹性回归中为许多变量提取公式

Extract Formula From RIDGE, LASSO, and Net Elastic Regression with Coefficients (R) for many variables

r

formula

linear-regression

glmnet

下面附上了我要提供的示例：

给出错误：