rfMt<-randomForest(x=xda[Idx,],y=yv[Idx],xtest=xda[-Idx,],ytest=yv[-Idx],ntree=1500)
用str() 显示模型结果 :
str(rfMt)
List of 17
$ call : language randomForest(x = xda[Idx, ], y = yv[Idx], xtest = xda[-Idx, ], ytest = yv[-Idx], ntree = 1500, corr.bias = TRUE)
$ type : chr "regression"
$ predicted : Named num [1:120] 1.22 2.06 1.58 1.91 1.45 ...
..- attr(*, "names")= chr [1:120] "69" "303" "22" "13" ...
$ mse : num [1:1500] 0.394 0.349 0.322 0.311 0.316 ...
$ rsq : num [1:1500]
................................
........................
$ y : num [1:120] 1.79 2.08 1.39 1.61 1.1 ...
$ test :List of 4
..$ predicted: Named num [1:83] 1.78 1.71 1.69 1.74 1.94 ...
.. ..- attr(*, "names")= chr [1:83] "5" "9" "12" "14" ...
..$ mse : num [1:1500] 0.392 0.292 0.303 0.316 0.305 ...
..$ rsq : num [1:1500] 0.345 0.512 0.493 0.471 0.49 ...
..$ proximity: NULL
$ inbag
注:红色部分即为plot(rfMat)输出部分,但不完美,因此,可自已编程绘制:
yda<-data.frame(oobmse=rfMt$mse,testmse=rfMt$test$mse)
> library(tidyr)
> yda$id<-1:1500
> str(yda)
'data.frame': 1500 obs. of 3 variables:
$ oobmse : num 0.394 0.349 0.322 0.311 0.316 ...
$ testmse: num 0.392 0.292 0.303 0.316 0.305 ...
$ id : int 1 2 3 4 5 6 7 8 9 10 ...
> yda<-gather(yda,key=Type,value=mse,-id)
> str(yda)
'data.frame': 3000 obs. of 3 variables:
$ id : int 1 2 3 4 5 6 7 8 9 10 ...
$ Type: Factor w/ 2 levels "oobmse","testmse": 1 1 1 1 1 1 1 1 1 1 ...
$ mse : num 0.394 0.349 0.322 0.311 0.316 ...
> library(ggplot2)
> ggplot(yda,aes(x=id,y=mse,colour=Type))+geom_line()
当然按自已的要求进一步美化!!!
当然mse 仅仅说明ntree 指标设置是否合理。test集的mse 则是一个预测性好坏的指标。
mse : (regression only) vector of mean square errors: sum of squared residuals divided by n.
|