作者：顾全，浙江大学软件工程硕士，现任桃树科技算法工程师
地址：https://github.com/ZJUguquan/OnlineRandomForest
参与：Cynthia
翻译：本文为天善智能编译，未经容许，禁止转载
介绍
Online Random Forest(ORF) 是由Amir Saffari等人最先提出。之后，Arthur Lui使用Python实现了算法。非常感谢他们的工作。在论文内容和Lui的算法的基础上，我通过R和R6包重构了代码。此外，ORF在此包中的实现，与randomForest结合，使它同时支持增量学习和批量学习，例如：在ORF的基础上构建树，然后通过ORF进行更新。通过这种方法，它将比以前快得多。
安装
if(!require(devtools)) install.packages("devtools")
devtools::install_github("ZJUguquan/OnlineRandomForest")
快速启动
最小举例:增量学习
library(OnlineRandomForest)param<-list('minSamples'=1,'minGain'=0.1,'numClasses'=3,'x.rng'=dataRange(iris[1:4]))orf<-ORF$new(param,numTrees=10)for(iin1:150)orf$update(iris[i,1:4], as.integer(iris[i,5]))cat("Mean depth of trees in the forest is:",orf$meanTreeDepth(),"\n")orf$forest[[2]]$draw()
## Mean depth of trees in the forest is: 3
## Root X4 < 1.21
## |----L: X3 < 2.38
## |----L: Leaf 1
## |----R: Leaf 2
## |----R: X4 < 2.15
## |----L: X1 < 4.92
## |----L: Leaf 3
## |----R: Leaf 3
## |----R: Leaf 3
分类举例
library(OnlineRandomForest)#data preparationdat<-iris;dat[,5]<-as.integer(dat[,5])x.rng<-dataRange(dat[1:4])param<-list('minSamples'=2,'minGain'=0.2,'numClasses'=3,'x.rng'=x.rng)ind.gen<-sample(1:150,30)#for generate ORFind.updt<-sample(setdiff(1:150,ind.gen),100)#for uodate ORFind.test<-setdiff(setdiff(1:150,ind.gen),ind.updt)#for test#construct ORF and updaterf<-randomForest::randomForest(factor(Species)~.,data=dat[ind.gen, ],maxnodes=2,ntree=100)orf<-ORF$new(param)orf$generateForest(rf,df.train=dat[ind.gen, ],y.col="Species")cat("Mean size of trees in the forest is:",orf$meanTreeSize(),"\n")
## Mean size of trees in the forest is: 3
for(iinind.updt) {orf$update(dat[i,1:4],dat[i,5])}cat("After update, mean size of trees in the forest is:",orf$meanTreeSize(),"\n")
## After update, mean size of trees in the forest is: 11.9
#predictorf$confusionMatrix(dat[ind.test,1:4],dat[ind.test,5],pretty=T)
##
##
## Cell Contents
## |-------------------------|
## | N |
## | N / Row Total |
## | N / Col Total |
## |-------------------------|
##
##
## Total Observations in Table: 20
##
##
## | actual
## prediction | 1 | 2 | 3 | Row Total |
## -------------|-----------|-----------|-----------|-----------|
## 1 | 4 | 0 | 0 | 4 |
## | 1.000 | 0.000 | 0.000 | 0.200 |
## | 1.000 | 0.000 | 0.000 | |
## -------------|-----------|-----------|-----------|-----------|
## 2 | 0 | 9 | 2 | 11 |
## | 0.000 | 0.818 | 0.182 | 0.550 |
## | 0.000 | 1.000 | 0.286 | |
## -------------|-----------|-----------|-----------|-----------|
## 3 | 0 | 0 | 5 | 5 |
## | 0.000 | 0.000 | 1.000 | 0.250 |
## | 0.000 | 0.000 | 0.714 | |
## -------------|-----------|-----------|-----------|-----------|
## Column Total | 4 | 9 | 7 | 20 |
## | 0.200 | 0.450 | 0.350 | |
## -------------|-----------|-----------|-----------|-----------|
##
##
#comparetable(predict(rf,newdata=dat[ind.test,])==dat[ind.test,5])
## FALSE TRUE
## 9 11
table(orf$predicts(X=dat[ind.test,])==dat[ind.test,5])
## FALSE TRUE
## 2 18
回归举例
#data preparationif(!require(ggplot2)) install.packages("ggplot2")data("diamonds",package="ggplot2")dat<-as.data.frame(diamonds[sample(1:53000,1000), c(1:6,8:10,7)])for(colinc("cut","color","clarity"))dat[[col]]<-as.integer(dat[[col]])#Don't forget thisx.rng<-dataRange(dat[1:9])param<-list('minSamples'=10,'minGain'=1,'maxDepth'=10,'x.rng'=x.rng)ind.gen<-sample(1:1000,800)ind.updt<-sample(setdiff(1:1000,ind.gen),100)ind.test<-setdiff(setdiff(1:1000,ind.gen),ind.updt)
#construct ORFrf<-randomForest::randomForest(price~.,data=dat[ind.gen, ],maxnodes=20,ntree=100)orf<-ORF$new(param)orf$generateForest(rf,df.train=dat[ind.gen, ],y.col="price")orf$meanTreeSize()
## [1] 39
#and updatefor(iinind.updt) {orf$update(dat[i,1:9],dat[i,10])}orf$meanTreeSize()
## [1] 105.7
#predict and compareif(!require(Metrics)) install.packages("Metrics")preds.rf<-predict(rf,newdata=dat[ind.test,])Metrics::rmse(preds.rf,dat$price[ind.test])
## [1] 988.8055
preds<-orf$predicts(dat[ind.test,1:9])Metrics::rmse(preds,dat$price[ind.test])#make progress
## [1] 869.9613
其他用途
在 Tree 类中
ta<-Tree$new("abc",NULL,NULL)tb<-Tree$new(1,Tree$new(36),Tree$new(3))tc<-Tree$new(89,tb,ta)tc$draw()
#update tctc$right$updateChildren(Tree$new("666"),Tree$new(999))tc$right$right$updateChildren(Tree$new("666"),Tree$new(999))tc$draw()
通过random Forest包配置一个Online random Tree,并升级
#data preparationlibrary(randomForest)dat1<-iris;dat1[,5]<-as.integer(dat1[,5])rf<-randomForest(factor(Species)~.,data=dat1,maxnodes=3)treemat1<-getTree(rf,1,labelVar=F)treemat1<-cbind(treemat1,node.ind=1:nrow(treemat1))x.rng1<-dataRange(dat1[1:4])param1<-list('minSamples'=5,'minGain'=0.1,'numClasses'=3,'x.rng'=x.rng1)ind.gen<-sample(1:150,50)#for generate ORTind.updt<-setdiff(1:150,ind.gen)#for update ORT#originort2<-ORT$new(param1)ort2$draw()
## Root 1
## Leaf 1
#generate a treeort2$generateTree(treemat1,df.node=dat1[ind.gen,])ort2$draw()
## Root X3 < 2.45
## |----L: Leaf 1
## |----R: X3 < 4.75
## |----L: Leaf 2
## |----R: Leaf 3
#update this treefor(iinind.updt) {ort2$update(dat1[i,1:4],dat1[i,5])}ort2$draw()
## Root X3 < 2.45
## |----L: Leaf 1
## |----R: X3 < 4.75
## |----L: Leaf 2
## |----R: X4 < 2.19
## |----L: X2 < 3.68
## |----L: X1 < 7.12
## |----L: X3 < 4.06
## |----L: Leaf 1
## |----R: Leaf 3
## |----R: Leaf 3
## |----R: Leaf 1
## |----R: Leaf 3