Tidymodels：执行 PCR 错误：无法对不存在的列进行子集化答案

【问题标题】：Tidymodels : problem performing PCR Error: Can't subset columns that don't existTidymodels：执行 PCR 错误：无法对不存在的列进行子集化
【发布时间】：2020-09-20 00:30:25
【问题描述】：

我正在尝试使用 tidymodels 进行 PCR，但是我一直遇到这个问题。我知道有一个类似的帖子，但那里的解决方案不适用于我的情况。

我的数据

library(AppliedPredictiveModeling)
data(solubility)

train = solTrainY %>% bind_cols(solTrainXtrans) %>% rename(solubility = ...1)

我的 PCR 分析

train %<>% mutate_all(., as.numeric) %>% glimpse()
tidy_rec = recipe(solubility ~ ., data = train) %>%
  step_corr(all_predictors(), threshold = 0.9) %>%
  step_pca(all_predictors(), num_comp = ncol(train)-1) %>% 
  prep()

tidy_rec %>% tidy(2) %>% select(terms) %>% distinct()

tidy_predata = tidy_rec %>% juice()

# Re-sampling
tidy_folds = vfold_cv(train, v = 10)

# Set model
tidy_rlm = linear_reg() %>% 
  set_mode("regression") %>% 
  set_engine("lm")

# Set workflow
tidy_wf = workflow() %>% 
  add_recipe(tidy_rec) %>% 
  add_model(tidy_rlm) 

# Fit model
tidy_fit = tidy_wf %>% 
  fit_resamples(tidy_folds) 

tidy_fit %>% collect_metrics()

错误

x Fold01: recipe: Error: Can't subset columns that don't exist.
x Columns `PC1`, `PC2`, `PC3`, `PC4`, and `PC5` don't exist.
x Fold02: recipe: Error: Can't subset columns that don't exist.
x Columns `PC1`, `PC2`, `PC3`, `PC4`, and `PC5` don't exist.
x Fold03: recipe: Error: Can't subset columns that don't exist.
x Columns `PC1`, `PC2`, `PC3`, `PC4`, and `PC5` don't exist.
x Fold04: recipe: Error: Can't subset columns that don't exist.
x Columns `PC1`, `PC2`, `PC3`, `PC4`, and `PC5` don't exist.
x Fold05: recipe: Error: Can't subset columns that don't exist.
x Columns `PC1`, `PC2`, `PC3`, `PC4`, and `PC5` don't exist.
x Fold06: recipe: Error: Can't subset columns that don't exist.
.
.
.

【问题讨论】：

这能回答你的问题吗？ Subsetting with $ when column may not exist

标签： r machine-learning pca tidymodels

【解决方案1】：

这是因为workflow 需要一个未准备好的配方规范。

因此，在您的代码中，从配方规范中删除 prep() 将消除错误。

tidy_rec <- recipe(solubility ~ ., data = train) %>%
  step_corr(all_predictors(), threshold = 0.9) %>%
  step_pca(all_predictors(), num_comp = ncol(train)-1) 
  # remove the prep() method

【讨论】：