Question

我正在尝试计算回归树预测的RMSE，但仍然会出错。

我已经加载了一个df，删除了所有空记录并将我的df拆分为训练和验证：

income.df <- read.csv("adult.data.csv")

income.df$Workclass <- gsub("[?]", NA, as.character(income.df$Workclass))
income.df$Workclass <- as.factor(income.df$Workclass)
income.df$occupation <- gsub("[?]", NA, as.character(income.df$occupation))
income.df$occupation <- as.factor(income.df$occupation)
income.df$native.country <- gsub("[?]", NA, as.character(income.df$native.country))
income.df$native.country <- as.factor(income.df$native.country)

income.sub.df <- na.omit(income.df)

levels(income.sub.df$income) <- c(1,0)

names(income.sub.df)[15] <- "isBelow50k"

train.index2 <- sample(income.sub.df$Age, dim(income.sub.df)[1]*0.6)
train2.df <- income.sub.df[train.index2, ]
valid2.df <- income.sub.df[-train.index2, ]

创建了一个rpart树

library(rpart)
library(rpart.plot)
library(forecast)

tr.income <- rpart(isBelow50k ~ ., data = train2.df)
prp(tr.income)
predrpart <- predict(tr.income, newdata = valid2.df)
accuracy(predrpart, valid2.df$isBelow50k)

我不断得到的错误是：

Error in accuracy(predrpart, valid2.df$isBelow50k) : 
  First argument should be a forecast object or a time series.

我已经尝试将其转换为矢量但似乎没有什么可行的。

Answer 1

您的问题是classification problem, not a regression problem。 RMSE（以及预测包中的accuracy函数）不用于分类问题。相反，我建议使用可以从confusion matrix派生的指标。

以下是修改代码以计算简单混淆矩阵的一个选项：

tr.income <- rpart(isBelow50k ~ ., data = train2.df)
prp(tr.income)
predrpart <- factor(ifelse(predict(tr.income, newdata = valid2.df)[,2] > .5, '>50K', '<=50K'))
table(predicted = predrpart, actual = valid2.df$isBelow50k)

计算回归树的准确性

1 个答案: