瞭解ROCR的performance（）函數返回的內容 - in R

Question

我很難理解ROCR包的performance（）函數返回的內容。讓我用一個可重複的例子來具體說明。我使用mpg數據集。我的代碼如下：

library(ROCR) 
library(ggplot2) 
library(data.table) 
library(caTools) 
data(mpg) 
setDT(mpg) 
mpg[year == 1999, Year99 := 1] 
mpg[year == 2008, Year99 := 0] 
table(mpg$Year99) 
# 0 1 
# 117 117 
split <- sample.split(mpg$Year99, SplitRatio = 0.75) 
mpg_train <- mpg[split, ] 
mpg_test <- mpg[!split, ] 
model <- glm(Year99 ~ displ, mpg_train, family = "binomial") 
summary(model) 
predict_mpg_test <- predict(model, type = "response", newdata = mpg_test) 
ROCR_mpg_test <- prediction(predict_mpg_test, mpg_test$Year99) 
performance(ROCR_mpg_test, "acc") 

#An object of class "performance" 
#Slot "x.name": 
# [1] "Cutoff" 

#Slot "y.name": 
# [1] "Accuracy" 

#Slot "alpha.name": 
# [1] "none" 

#Slot "x.values": 
# [[1]] 
#49  55  56  45  47  53  51  57  46  13  39  37  58 
#Inf 0.5983963 0.5926422 0.5868625 0.5752326 0.5635187 0.5576343 0.5458183 0.5398901 0.5280013 0.5220441 0.5101127 0.4981697 0.4921981 
#17  44  31  32  33  50  34  40  24  21  12 
#0.4802634 0.4683511 0.4564748 0.4446478 0.4328831 0.4270282 0.4095919 0.3923800 0.3866994 0.3698468 0.3265163 


#Slot "y.values": 
# [[1]] 
#[1] 0.5000000 0.5172414 0.5344828 0.5344828 0.5517241 0.5344828 0.4827586 0.5000000 0.5862069 0.6206897 0.6034483 0.6206897 0.5862069 
#[14] 0.5689655 0.5517241 0.5689655 0.5517241 0.5344828 0.5517241 0.5172414 0.5344828 0.4655172 0.4827586 0.4827586 0.5000000 


#Slot "alpha.values": 
# list() 

我的問題是：

1. 什麼是4行下槽 「x.values」 中列出的數字？
2. 插槽「y.values」下面列出的2行數字是什麼？
3. 是否有可能向ROCR函數傳遞一系列截斷值 - 例如， cutoff = seq（0.05,0.95,0.05） - 然後返回給定的度量值 - 例如，準確度---對於每個截止水平？

您的建議將不勝感激。

Answer 1

（1）在x.values插槽中可以找到截止點。
截止值的該載體含有所述組由所述模型預測的概率的唯一值：

prf <- performance(ROCR_mpg_test, "acc") 

cutoffs <- prf@x.values[[1]] 
pred.probs <- sort(unique(predict_mpg_test), decreasing=T) 
all(cutoffs[-1] == pred.probs) 
# [1] TRUE 

（2）在y.values時隙存在用於各個切割的準確度。

accuracies1 <- prf@y.values[[1]] 

# Example. Calculate accuracy for the 3rd cutoff 
(tbl <- table(predict_mpg_test>= cutoffs[3], mpg_test$Year99)) 
#   0 1 
# FALSE 28 25 
# TRUE 1 4 
sum(diag(tbl))/sum(tbl) 
# [1] 0.5517241 
accuracies1[3] 
# [1] 0.5517241 

# Calcuate the accuracies for each cutoff 
calc_accur <- function(cutoff, pred_prob, response_var) { 
    confusion_matrix <- table(pred_prob >= cutoff, response_var) 
    sum(diag(confusion_matrix))/sum(confusion_matrix) 
} 

accuracies2 <- sapply(cutoffs, calc_accur, 
     pred_prob=predict_mpg_test, response_var=mpg_test$Year99) 

all(accuracies1==accuracies2) 
# [1] TRUE 

（3）使用在（2）和sapply有可能給出的calc_accur函數來傳遞截斷的序列，並且計算出相應的精度。
例如：

seq_cut <- seq(0.3, 0.6, length.out=10) 
sapply(seq_cut, calc_accur, 
     pred_prob=predict_mpg_test, response_var=mpg_test$Year99) 

# [1] 0.5000000 0.5000000 0.5000000 0.5172414 0.5517241 0.5862069 0.6551724 0.6379310 
# [9] 0.5517241 0.5000000