人大经济论坛 › 标签 › Party › 相关日志

标签: Party经管大学堂：名校名师名课

相关日志

分享转一个code,有时间测试一下: 唐伯小猫 2014-4-26 05:19; library(rpart) #Popular decision tree algorithm library(rattle) #Fancy tree plot, nice graphical interface library(rpart.plot) #Enhanced tree plots library(RColorBrewer) #Color selection for fancy tree plot library(party) #Alternative decision tree algorithm library(partykit) #Convert rpart object to BinaryTree library(RWeka) #Weka decision tree J48 library(evtree) #Evolutionary Algorithm, builds the tree from the bottom up library(randomForest) library(doParallel) library(CHAID) #Chi-squared automatic interaction detection tree library(tree) library(caret) ls(package:party) #list functions in package party #Data Prep data(weather) dsname - “weather” target - “RainTomorrow” risk - “RISK_MM” ds - get(dsname) vars - colnames(ds) (ignore - vars ) vars - setdiff(vars, ignore) (inputs - setdiff(vars, target)) (nobs - nrow(ds)) dim(ds ) (form - formula(paste(target, “~ .”))) set.seed(1426) length(train - sample(nobs, 0.7*nobs)) length(test - setdiff(seq_len(nobs), train)) dim(ds) head(ds) tail(ds) summary(ds) str(ds) #——————————————————————- # Basic Scatterplot Matrix pairs(paste(“~”, paste(vars, collapse=’+'), sep=”),data=ds, main=”Simple Scatterplot Matrix”) pairs(~MinTemp+MaxTemp+Rainfall+Evaporation, data =ds, main=”Simple Scatterplot Matrix”) histogram(ds$MinTemp, breaks=20, col=”blue”) #——————————————————————- #Rpart Tree library(rpart) model - rpart(formula=form, data=ds ) model summary(model) printcp(model) #printcp for rpart objects plotcp(model) plot(model) text(model) fancyRpartPlot(model) prp(model) prp(model, type=2, extra=104, nn=TRUE, fallen.leaves=TRUE, faclen=0, varlen=0, shadow.col=”grey”, branch.lty=3) pred - predict(model, newdata=ds , type=”class”) #na.action = na.pass pred.prob - predict(model, newdata=ds , type=”prob”) #Check for na in the data, remove rows, if there are NA’s, rpart will use surrogate splits. table(is.na(ds)) ds.complete - ds (nobs - nrow(ds.complete)) set.seed(1426) length(train.complete - sample(nobs, 0.7*nobs)) length(test.complete - setdiff(seq_len(nobs), train.complete)) #Prune tree model$cptable ),”CP”] #want the first minimum model - rpart(formula=form, data=ds , cp=0) printcp(model) prune - prune(model, cp=.01) printcp(prune) #——————————————————————- #Party Tree install.packages(“partykit”, repos=”http://R-Forge.R-project.org”) library(partykit) class(model) plot(as.party(model)) #——————————————————————- #tree model - tree(formula=form, data=ds ) summary(model) #——————————————————————- #Conditional Inference Tree model - ctree(formula=form, data=ds ) model plot(model) pred - predict(model, newdata=ds ) pred.prob - predict(model, newdata=ds , type=”prob”) #Try this for class predictions: library(caret) confusionMatrix(pred, ds ) mc - table(pred, ds ) err - 1.0 – (mc + mc ) / sum(mc) #resubstitution error rate #For class probabilities: probs - treeresponse(model, newdata=test) pred - do.call(rbind, as.list(pred)) summary(pred) #For a roc curve: library(ROCR) roc - prediction(pred , ds ) #noquote(paste(“test$”, target, sep=”)) plot(performance(roc, measure=”tpr”, x.measure=”fpr”), colorize=TRUE) #For a lift curve: plot(performance(roc, measure=”lift”, x.measure=”rpp”), colorize=TRUE) #Sensitivity/specificity curve and precision/recall curve: #sensitivity(i.e True Positives/Actual Positives) and specifcity(i.e True Negatives/Actual Negatives) plot(performance(roc, measure=”sens”, x.measure=”spec”), colorize=TRUE) plot(performance(roc, measure=”prec”, x.measure=”rec”), colorize=TRUE) #Here’s an example of using 10-fold cross-validation to evaluation your model library(doParallel) registerDoParallel(cores=2) model - train(ds , ds , method=’rpart’, tuneLength=10) #cross validation #example n - nrow(ds) #nobs K - 10 #for 10 validation cross sections taille - n%/%K set.seed(5) alea - runif(n) rang - rank(alea) bloc - (rang-1)%/%taille +1 bloc - as.factor(bloc) print(summary(bloc)) all.err - numeric(0) for(k in 1:K){ model - rpart(formula=form, data = ds , method=”class”) pred - predict(model, newdata=ds , type=”class”) mc - table(ds ,pred) err - 1.0 – (mc +mc ) / sum(mc) all.err - rbind(all.err,err) } print(all.err) (err.cv - mean(all.err)) #——————————————————————- #Weka Decision Tree model - J48(formula=form, data=ds ) model predict - predict(model, newdata=ds ) predict.prob - predict(model, newdata=ds , type=”prob”) #——————————————————————- #Evolutionary Trees target - “RainTomorrow” model - evtree(formula=form, data=ds ) model plot(model) predict - predict(model, newdata=ds ) predict.prob - predict(model, newdata=ds , type=”prob”) #——————————————————————- #Random Forest from library(randomForest) table(is.na(ds)) table(is.na(ds.complete)) setnum - colnames(ds.complete) #subset(ds, select=-c(Humidity3pm, Humidity9am, Cloud9am, Cloud3pm)) ds.complete - lapply(ds.complete , function(x) as.numeric(x)) ds.complete$Humidity3pm - as.numeric(ds.complete$Humidity3pm) ds.complete$Humidity9am - as.numeric(ds.complete$Humidity9am) begTime - Sys.time() set.seed(1426) model - randomForest(formula=form, data=ds.complete ) runTime - Sys.time()-begTime runTime #Time difference of 0.3833725 secs begTime - Sys.time() set.seed(1426) model - randomForest(formula=form, data=ds.complete , ntree=500, replace = FALSE, sampsize = .632*.7*nrow(ds), na.action=na.omit) runTime - Sys.time()-begTime runTime #Time difference of 0.2392061 secs model str(model) pred - predict(model, newdata=ds.complete ) #Random Forest in parallel library(doParallel) ntree = 500 numCore = 4 rep - 125 # tree / numCore registerDoParallel(cores=numCore) begTime - Sys.time() set.seed(1426) rf - foreach(ntree=rep(rep, numCore), .combine=combine, .packages=’randomForest’) %dopar% randomForest(formula=form, data=ds.complete , ntree=ntree, mtry=6, importance=TRUE, na.action=na.roughfix, #can also use na.action = na.omit replace=FALSE) runTime - Sys.time()-begTime runTime #Time difference of 0.1990662 secs importance(model) importance(rf) pred - predict(rf, newdata=ds.complete ) confusionMatrix(pred, ds.complete ) #Random Forest from library(party) model - cforest(formula=form, data=ds.complete ) #Factor Levels id - which(!(ds$var.name %in% levels(ds$var.name))) ds$var.name - NA #——————————————————————- #Regression Trees – changing target and vars target - “RISK_MM” vars - c(inputs, target) form - formula(paste(target, “~ .”)) (model - rpart(formula=form, data=ds )) plot(model) text(model) prp(model, type=2, extra=101, nn=TRUE, fallen.leaves=TRUE, faclen=0, varlen=0, shadow.col=”grey”, branch.lty=3) rsq.rpart(model) library(Metrics) pred - predict(model, newdata=ds ) err - rmsle(ds , pred) #compare probabilities not class #——————————————————————- #Chaid Tree – new data set data(“BreastCancer”, package = “mlbench”) sapply(BreastCancer, function(x) is.factor(x)) b_chaid - chaid(Class ~ Cl.thickness + Cell.size + Cell.shape + + Marg.adhesion + Epith.c.size + Bare.nuclei + + Bl.cromatin + Normal.nucleoli + Mitoses, data = BreastCancer) plot(b_chaid) #——————————————————————- #List functions from a package ls(package:rpart) #save plots as pdf pdf(“plot.pdf”) fancyRpartPlot(model) dev.off(); 0 个评论

京ICP备16021002号-2 京B2-20170662号京公网安备 11010802022788号论坛法律顾问：王进律师知识产权保护声明免责及隐私声明