是否需要规范化数据以使配方中的自然样条萌发

问题描述

我使用自然样条曲线拟合模型，但不确定使用BoxCox并居中并缩放预测变量是否有优势。阶梯自然样条线是否执行转换？在自然样条之前对预测变量进行归一化有一些优势吗？

library(tidyverse)
library(tidymodels)

car <- read_csv('vw.csv')
str(car)
## ---- Split data -----------------------

split <- initial_split(car,prop = 0.80,strata = 'price')
car_train <- training(split)
car_test <- testing(split)


## ---- Recipe --------------------------

rec <- recipe(price ~ .,data = car_train) %>% 
  step_mutate(
    tax = log(tax + 1)
  ) %>% 
  step_ns(mpg,mileage,enginesize,year,deg_free = 3) %>% 
  step_dummy(all_nominal()) 


## -- Model ---------------------------------

model_lasso <- linear_reg(mode = 'regression',penalty = tune(),mixture = tune()) %>% 
               set_engine('glmnet')


## --- Workflow -----------------------------

work01 <- workflow() %>% 
  add_recipe(rec) %>% 
  add_model(model_lasso)

## --- Foldes -------------------------------

folds <- vfold_cv(car_train,v = 10,strata = 'price')

## --- tune ---------------------------------

grid01 <- grid_latin_hypercube(parameters(model_lasso),size = 10)

tune01 <- tune_grid(
  work01,resamples = folds,grid = grid01,metrics = metric_set(rmse,rsq)
)

## --- Show_best ---------------------------------

show_best(tune01)
best01 <- select_best(tune01)

## --- Test --------------------------------------

test01 <- work01 %>% 
  finalize_workflow(best01) %>%
  last_fit(split)

test01 %>% collect_metrics()

´´´

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）