1

我正在使用 mlr 和其他软件包进行生存分析。在 mlr 中，我使用 surv.rpart 和 surv.glmboost。我还使用原始包 rpart 和 mboost 来执行此操作。我发现他们的结果是不同的。这是一个例子：

> myData2 <- data.frame(DaySum=c(3,2,1,6,3,2,2,5,2,7,2),
                        DaysDiff=c(24,4,5,12,3,31,131,6,35,18,19),
                        Status='TRUE')
> myData2$Status <- as.logical(myData2$Status)
> myTrain <- c(1:(nrow(myData2)-1))
> myTest <- nrow(myData2)

当我在 mlr 中使用 surv.rpart 时，结果是：

> surv.task <- makeSurvTask(data=myData2,target=c('DaysDiff','Status'))
> surv.lrn <- makeLearner("surv.rpart")
> mod <- train(learner=surv.lrn,task=surv.task,subset=myTrain)
> surv.pred <- predict(mod,task=surv.task,subset=myTest)
> surv.pred
Prediction: 1 observations
predict.type: response
threshold: 
time: 0.00
   id truth.time truth.event response
11 11         19        TRUE        1

如果我使用原始的 rpart 包，结果是：

> train <- myData2[1:(nrow(myData2)-1),]
> test <- myData2[nrow(myData2),]
> fit <- rpart(DaysDiff~DaySum,data=train)
> predict(fit,newdata=test)
[1] 26.9

为什么我得到了两个不同的结果？看起来 rpart 包直接给了我想要的结果，而来自 mlr 的结果有某种转换。当我使用 surv.glmboost 时也会发生同样的事情：

> surv.task <- makeSurvTask(data=myData2,target=c('DaysDiff','Status'))
Warning messages:
1: Unknown or uninitialised column: 'Weibull'. 
2: Unknown or uninitialised column: 'Cox'. 
3: Unknown or uninitialised column: 'Month2'. 
4: Unknown or uninitialised column: 'Month2'. 
5: Unknown or uninitialised column: 'Month'. 
6: Unknown or uninitialised column: 'Month'. 
7: Unknown or uninitialised column: 'MonthsDiff'. 
8: Unknown or uninitialised column: 'Weibull'. 
9: Unknown or uninitialised column: 'Cox'. 
> surv.lrn <- makeLearner("surv.glmboost")
> mod <- train(learner=surv.lrn,task=surv.task,subset=myTrain)
Warning message:
In names(data) != all.vars(formula[[2]]) :
  longer object length is not a multiple of shorter object length
> surv.pred <- predict(mod,task=surv.task,subset=myTest)
> surv.pred
Prediction: 1 observations
predict.type: response
threshold: 
time: 0.00
   id truth.time truth.event   response
11 11         19        TRUE -0.1946239

这是使用 mboost 包的结果：

> train <- myData2[1:(nrow(myData2)-1),]
Warning messages:
1: Unknown or uninitialised column: 'Weibull'. 
2: Unknown or uninitialised column: 'Cox'. 
3: Unknown or uninitialised column: 'Month2'. 
4: Unknown or uninitialised column: 'Month2'. 
5: Unknown or uninitialised column: 'Month'. 
6: Unknown or uninitialised column: 'Month'. 
7: Unknown or uninitialised column: 'MonthsDiff'. 
8: Unknown or uninitialised column: 'Weibull'. 
9: Unknown or uninitialised column: 'Cox'. 
> test <- myData2[nrow(myData2),]
> fit <- glmboost(DaysDiff~DaySum,data=train)
> predict(fit,newdata=test)
         [,1]
[1,] 33.08294

这是我到目前为止发现的。这可能发生在 surv.cforest 等其他函数上。我的问题是：为什么会这样？以及如何在使用 mlr 包时获得 rpart 和 mboost 等结果？

4

1 回答 1

1

你的问题是，你不是用 rpart 和 glmboost 拟合生存模型，而是一个简单的回归模型。

在 rpart 中拟合生存模型如下所示：

fit = rpart(Surv(DaysDiff, event = Status) ~ DaySum,data=train, method = "exp")
predict(fit,newdata=test)

所以完整的比较代码给出了相同的结果（每个预测 1）：

library(mlr)
myData2 = data.frame(DaySum=c(3,2,1,6,3,2,2,5,2,7,2),
  DaysDiff=c(24,4,5,12,3,31,131,6,35,18,19),
  Status='TRUE')
myData2$Status = as.logical(myData2$Status)
train = myData2[1:(nrow(myData2)-1),]
test = myData2[nrow(myData2),]
surv.task = makeSurvTask(data=train,target=c('DaysDiff','Status'))
surv.lrn = makeLearner("surv.rpart")
mod = train(learner=surv.lrn,task=surv.task,subset=myTrain)
surv.pred = predict(mod,newdata = test)
surv.pred
library(rpart)
library(survival)
fit = rpart(Surv(DaysDiff, event = Status) ~ DaySum,data=train, method = "exp")
predict(fit,newdata=test)
于 2017-08-16T09:54:30.633 回答