Question

I'm creating a decision tree with the R rpart package based on x number of variables and a dataframe:

fit<-rpart(y~x1+x2+x3+x4,data=(mydataframe),
  control=rpart.control(minsplit = 20, minbucket = 0, cp=.01))

But instead of using the entire dataframe, I have four or five subsets of data that are factors, let's say separated out by x4. How can I run decision trees on all of these factors at once instead of having to call subsets of the data again and again?

Based on a search of SO, it looks like either BY or ddply might be the right choice. Here's what I've tried for ddply:

fit<-ddply(mydataframe, dataframe$x4, function (df)  
    rpart(y~x1+x2+x3+x4,data=(df), 
    control=rpart.control(minsplit = 20, minbucket = 0, cp=.01)))

but what I'm getting back is:

Error in eval(expr, envir, enclos) : object 'x4value' not found

where x4value is one of the variable values I'd like to split out by. So I have a column of values:

x4
BucketName1
BucketName2
BucketName3
BucketName4

str(mydataframe) shows that $x4 is a : Factor w/ 8 levels and no symbols.

Additionally, I ran mydataframe = na.omit(dataframe) at the very beginning to avoid nulls.

Possible issues I've already troubleshooted:

Variable name
Object not found referencing an object that doesn't exist (not the case since my column name exists)
- The values of BucketName1, 2, 3, and 4 not allowed to start with a number, which I fixed.

The rpart bit runs fine when I run it manually as such:

mydataframe<-subset(trainData, x4=="BucketName1")

fit<-rpart(y~x1+x2+x3+x4,data=(mydataframe), 
    control=rpart.control(minsplit = 20, minbucket = 0, cp=.01))

but borks whenever I try to loop through all subsets using ddply.

Complete reproducible sample code:

mydataframe<-data.frame  ( x1=sample(1:10),
                           x2=sample(1:10),
                           x3=sample(1:10),
                           x4= sample(letters[1:4], 20, replace = TRUE))
str(mydataframe)

fit<-ddply(mydataframe, mydataframe$x4, function (df)
    rpart(y~x1+x2+x3+x4,data=(df), control=rpart.control(minsplit = 20,      minbucket = 0, cp=.01)))

Output:

str(mydataframe) 'data.frame':  20 obs. of  4 variables:  $ x1: int  1 6 8 4 7 9 3 2 10 5 ...  $ x2: int  9 4 5 8 6 3 7 10 2 1 ...  $ x3: int 2 6 5 3 1 4 9 7 10 8 ...  $ x4: Factor w/ 4 levels "a","b","c","d": 4 4 3 2 3 4 3 3 1 3 ...
> fit<-ddply(mydataframe, mydataframe$x4, function (df) rpart(y~x1+x2+x3+x4,data=(df), control=rpart.control(minsplit = 20, minbucket = 0, cp=.01))) Error in eval(expr, envir, enclos) : object 'd' not found

Answer 1

你想用你的代码做两件事：

使用dlply代替ddply，因为您需要一个rpart对象列表而不是（？）的数据框。如果您想显示原始数据的预测值，ddply会很有用，因为可以将其格式化为数据框。
在.(x4)中使用dataframe$x4代替dlply。使用后者将产生不可预测的结果。

此外，在您的示例中，您应指定y值并从....之后删除x4

Answer 2

您将错误的值传递给dplyr() .variables=参数。您应该传递引用的变量名称，公式或变量名称的字符向量。由于您正在将mydataframe$v4传递给一个角色，并且它正在寻找该列中的所有值，就好像它们是变量一样。

这是电话应该是什么样子

fit<-ddply(mydataframe, ~x4, function (df)
    rpart(y~x1+x2+x3+x4,data=(df), control=rpart.control(minsplit = 20, minbucket = 0, cp=.01)))

或

fit<-ddply(mydataframe, .(x4), function (df)
    rpart(y~x1+x2+x3+x4,data=(df), control=rpart.control(minsplit = 20, minbucket = 0, cp=.01)))

或

fit<-ddply(mydataframe, "x4", function (df)
    rpart(y~x1+x2+x3+x4,data=(df), control=rpart.control(minsplit = 20,  minbucket = 0, cp=.01)))

Answer 3

如果您对plyr不熟悉，也可以使用基本R功能执行此操作。

splitData = split(mydataframe, mydataframe$x4)

getModel = function(df) {
    fit <- rpart(y~x1+x2+x3+x4+xN....,data=df, 
        control=rpart.control(minsplit = 20, minbucket = 0, cp=.01)))
    return(fit)
}

models = lapply(splitData, getModel)

您也可以使用dplyr而不是plyr执行此操作。

mydataframe %>% group_by(x4) %>%
   do(model = getModel(.))

Running rpart over multiple subsets of a data frame

3 个答案: