如何修复 R 中的此错误：[.data.frame(newdata, , object$method$center, drop = FALSE) : undefined columns selected

Question

我正在尝试从论文中重新创建一个随机森林模型，但代码似乎不起作用，我只是在学习 R，这让我很头疼，但我会尽力解释。

论文的源代码可以在这里找到：[(https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0181347.s002&type=supplementary)]

本文提供两个数据集：训练和测试，然后为每个数据集创建两个子集（数据的 head() 参见文本底部。Data can be found in supplementary paper here

（应该可以直接复制到.csv）代码如下：

sink("test.txt", split=TRUE)
print("#data process")
data_bin_train<-read.csv("training.csv", head=TRUE)
names(data_bin_train)
data_bin_test<-read.csv("test.csv", head=TRUE)
names(data_bin_test)
dspt_bin_train<-subset(data_bin_train,select=c(-Deamidation))
dspt_bin_test<-subset(data_bin_test,select=c(-Deamidation))
class_bin_train<-subset(data_bin_train, select=c(Deamidation))
class_bin_test<-subset(data_bin_test, select=c(Deamidation))

library("caret")
library("ROCR")
library("pROC")
fitControl <- trainControl(method = "CV",number = 10,returnResamp = "all", verboseIter = FALSE, classProbs = TRUE)
set.seed(2)

这个位工作正常。然后下一段代码是我得到错误的地方：

library("randomForest")
print("#Random Forest binary class via caret (randomForest)")
caret_rf_bin_randomf_cv10 <- train(Deamidation~., data=data_bin_train, method = "rf", preProcess = c("center", "scale"), tuneLength = 10, trControl = fitControl)
caret_rf_bin_randomf_cv10
varImp(caret_rf_bin_randomf_cv10)

rf_bin_Preds <- extractPrediction(list(caret_rf_bin_randomf_cv10),testX=dspt_bin_test[,1:13], testY=class_bin_test[,1]) 

[.data.frame(newdata, , object$method$center, drop = FALSE) 中的错误：选择了未定义的列` 任何帮助都会很棒！该论文使用 R v 3.1.1 caret_6.0-35，而我正在运行两者的更新版本，这是我认为错误来自的地方，但我不确定如何修复它，或者说实话是什么错误甚至是。

谢谢

TinoMass

下面是两个数据集的`sessionInfo() 和 Head()

R version 3.5.3 (2019-03-11)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 18362)

Matrix products: default

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] randomForest_4.6-14 pROC_1.15.3         ROCR_1.0-7          gplots_3.0.1.1      caret_6.0-84       
[6] ggplot2_3.2.1       lattice_0.20-38    

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.3         pillar_1.4.3       compiler_3.5.3     gower_0.2.1        plyr_1.8.5         bitops_1.0-6      
 [7] iterators_1.0.12   class_7.3-15       tools_3.5.3        rpart_4.1-13       ipred_0.9-9        lubridate_1.7.4   
[13] lifecycle_0.1.0    tibble_2.1.3       nlme_3.1-137       gtable_0.3.0       pkgconfig_2.0.3    rlang_0.4.2       
[19] Matrix_1.2-15      foreach_1.4.7      rstudioapi_0.10    prodlim_2019.11.13 e1071_1.7-3        withr_2.1.2       
[25] stringr_1.4.0      dplyr_0.8.3        caTools_1.17.1.3   gtools_3.8.1       generics_0.0.2     recipes_0.1.8     
[31] stats4_3.5.3       grid_3.5.3         nnet_7.3-12        tidyselect_0.2.5   data.table_1.12.8  glue_1.3.1        
[37] R6_2.4.1           survival_2.43-3    gdata_2.18.0       lava_1.6.6         reshape2_1.4.3     purrr_0.3.3       
[43] magrittr_1.5       ModelMetrics_1.2.2 scales_1.1.0       codetools_0.2-16   MASS_7.3-51.1      splines_3.5.3     
[49] assertthat_0.2.1   timeDate_3043.102  colorspace_1.4-1   KernSmooth_2.23-15 stringi_1.4.3      lazyeval_0.2.2    
[55] munsell_0.5.0      crayon_1.3.4     

training.txt

PDB   `Residue #` `AA following A… attack_distance Half_life norm_B_factor_C norm_B_factor_CA norm_B_factor_CB norm_B_factor_CG
  <chr>       <dbl> <chr>                      <dbl>     <dbl>           <dbl>            <dbl>            <dbl>            <dbl>
1 11BG           67 GLY                         3.84      1.02           1.46             1.46             1.36             1.38 
2 11BG           17 SER                         4.81     11.8            0.692            0.706            1.18             1.62 
3 11BG           71 CYS                         4.11     55.5            0.174            0.481            0.574            0.782
4 11BG           44 THR                         3.33     49.9           -1.24            -1.30            -1.35            -1.52 
5 11BG           94 CYS                         4.97     60              1.41             1.64             1.92             2.15 
6 11BG           27 LEU                         4.52    119             -0.898           -0.905           -0.820           -0.604

test.txt

PDB   `Residue #` `AA following A… attack_distance Half_life norm_B_factor_C norm_B_factor_CA norm_B_factor_CB norm_B_factor_CG
 <chr>       <dbl> <chr>                      <dbl>     <dbl>           <dbl>            <dbl>            <dbl>            <dbl>
1 1ACC          713 GLY                         3.69      1.45           3.17             3.35              3.63             4.06
2 1ACC          719 GLY                         4.64      1.04           0.688            0.865             1.42             1.83
3 1ACC           28 PHE                         4.81     72.4            1.03             1.06              1.58             1.95
4 1ACC           52 ILE                         4.73    279              0.944            1.13              1.29             1.46
5 1ACC           85 HIS                         3.60      9.7            0.780            0.800             1.16             1.57
6 1ACC          104 LYS                         4.51     55.5            2.22             2.47              2.69             2.91

Answer 1

问题出在两个功能上：

“PDB” - 根本不应该使用，因为它只是序列加入。

“AA.following.Asn” - 测试集不包含所有级别作为训练集。

levels(data_bin_train[,3])
#output
 [1] "ALA" "ARG" "ASP" "CYS" "GLU" "GLY" "HIS" "ILE" "LEU" "LYS" "MET" "PHE" "PRO" "SER" "THR" "TRP" "TYR" "VAL"

levels(data_bin_test[,3])
#output
 [1] "ALA" "ARG" "ASP" "GLU" "GLY" "HIS" "ILE" "LEU" "LYS" "PHE" "PRO" "SER" "THR" "TRP" "TYR" "VAL"

如果从集合中省略这两个特性，代码运行并产生合理的结果

library(caret)
fitControl <- trainControl(method = "CV",
                           number = 10,
                           returnResamp = "all",
                           verboseIter = FALSE,
                           classProbs = TRUE)
set.seed(2)
caret_rf_bin_randomf_cv10 <- train(Deamidation~.,
                                   data = data_bin_train[,-c(1,3)],
                                   method = "rf",
                                   preProcess = c("center", "scale"),
                                   tuneLength = 10,
                                   trControl = fitControl)

rf_bin_Preds <- extractPrediction(list(caret_rf_bin_randomf_cv10),
                                  testX = data_bin_test[,-c(1,3)],
                                  testY = class_bin_test$Deamidation) 

head(rf_bin_Preds)
#output
  obs pred model dataType  object
1 Yes  Yes    rf Training Object1
2  No   No    rf Training Object1
3  No   No    rf Training Object1
4  No   No    rf Training Object1
5  No   No    rf Training Object1
6  No   No    rf Training Object1