Caret 预测目标变量 nrow() 为 Null

Question

df:

library(caret)

a = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
b = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
c = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
d = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
e = c(1, 0, 1, 0, 0, 0, 1, 1, 1, 1)

#df1
df1 = data.frame(a,b,c,d,e)
#df2
df2 = data.frame(a,b,c,d,e)

Caret Log-红色模型：

df1$e <- as.factor(df1$e)
df2$e <- as.factor(df2$e)

# define training control
train_control <- trainControl(method = "cv", number = 5)

# train the model on training set
model <- train(e ~ .,
               data = df1,
               trControl = train_control,
               method = "glm",
               family=binomial())

# logistic <- glm(WonLost ~ . -PANum, data=train, family="binomial")
df2$predict <- caret::predict.train(model, newdata=df2,type = "prob")


nrow(df2$predict)
nrow(df2$e)

为什么 nrow(df2$e) 为零？我根据之前遇到的错误将目标变量更改为一个因子，但这似乎导致了我当前的问题。

Warning messages: 1: In train.default(x, y, weights = w, ...) : You are trying to do regression and your outcome only has two possible values Are you trying to do classification? If so, use a 2 level factor as your outcome column.

Answer 1

有时 caret 对变量敏感，即使您的 glm logit 模型在回归或分类方面存在问题也是如此。我学到的一个建议是将目标变量重新编码为 Yes/No。另外，请注意插入符号的预测被添加为 df2 中的新数据帧，这就是为什么 nrow() 有效而 e 只是一个向量，因此您必须使用 length() 或NROW()。这里的代码：

library(caret)
#Vectors
a = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
b = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
c = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
d = c("aa", "bb", "cc", "aa", "aa", "aa", "bb", "cc", "bb", "bb") 
e = c(1, 0, 1, 0, 0, 0, 1, 1, 1, 1)

#df1
df1 = data.frame(a,b,c,d,e)
#df2
df2 = data.frame(a,b,c,d,e)
#Format
df1$e[df1$e==1] <- 'Yes'
df1$e[df1$e==0] <- 'No'
df2$e[df2$e==1] <- 'Yes'
df2$e[df2$e==0] <- 'No'

# define training control
train_control <- trainControl(method = "cv", number = 5)

# train the model on training set
model <- train(e ~ .,
               data = df1,
               trControl = train_control,
               method = "glm",
               family=binomial())

#Predict
df2$predict <- caret::predict.train(model, newdata=df2,type = "prob")
#Checks
nrow(df2$predict)
NROW(df2$e)
length(df2$e)

输出：

df2
    a  b  c  d   e   predict.No predict.Yes
1  aa aa aa aa Yes 7.500000e-01        0.25
2  bb bb bb bb  No 2.500000e-01        0.75
3  cc cc cc cc Yes 8.646869e-09        1.00
4  aa aa aa aa  No 7.500000e-01        0.25
5  aa aa aa aa  No 7.500000e-01        0.25
6  aa aa aa aa  No 7.500000e-01        0.25
7  bb bb bb bb Yes 2.500000e-01        0.75
8  cc cc cc cc Yes 8.646869e-09        1.00
9  bb bb bb bb Yes 2.500000e-01        0.75
10 bb bb bb bb Yes 2.500000e-01        0.75

nrow(df2$predict)
[1] 10
NROW(df2$e)
[1] 10
length(df2$e)
[1] 10

Caret 预测目标变量 nrow() 为 Null

Caret Predict Target Variable nrow() is Null

r

r-caret