DSASD
[#permalink]
09 Mar 2016, 10:12
install.packages("gbm")
require("gbm")
install.packages("dplyr")
require("dplyr")
install.packages("caret")
require("caret")
install.packages("ROCR")
require("ROCR")
train = read.csv("//namcrdnjdfs001/NASIRV01CCAR/Groups/Decision-RiskMGT/Risk/Citi Risk Modeling/Cards Corp/Projects/Custom Model Development/ECM/Alternative_Techniques/kaggle/train.csv",header = TRUE)
head(train)
indep= select(train
,-ID)
head(indep)
gbm_shrinkage = 0.1
gbm_depth = 4
gbm_minobs = 100
gbm_ntrees = 130
gbm_model =gbm(formula = TARGET ~.
,data = indep
,distribution = "adaboost",bag.fraction = 0.5, shrinkage =gbm_shrinkage
,n.trees = gbm_ntrees
,interaction.depth= gbm_depth,n.minobsinnode = gbm_minobs, verbose = TRUE
,cv.folds=5)
warnings()
summary(gbm_model)
gbm.perf(gbm_model,method="cv")
indep$pred = predict.gbm(object=gbm_model,newdata=indep,gbm_ntrees,type="response")
prob = predict.gbm(object=gbm_model,newdata=indep,gbm_ntrees, type= "response")
D_Pred <- prediction(prob,indep$TARGET)
D_Perf <- performance(D_Pred,"tpr","fpr")
D_KS <- max(attr(D_Perf,'y.values')[[1]]-attr(D_Perf,'x.values')[[1]])
D_perf1 <- performance(D_Pred,"auc")
D_AUC <- attr(D_perf1,'y.values')[[1]]
D_Gini <- 2*D_AUC-1
install.packages("randomForest")
require("randomForest")
install.packages("dplyr")
require("dplyr")
install.packages("caret")
require("caret")
install.packages("ROCR")
require("ROCR")
train = read.csv("//namcrdnjdfs001/NASIRV01CCAR/Groups/Decision-RiskMGT/Risk/Citi Risk Modeling/Cards Corp/Projects/Custom Model Development/ECM/Alternative_Techniques/snapshot/FICODEV_SAMPLE.csv",header = TRUE)
valid = read.csv("//namcrdnjdfs001/NASIRV01CCAR/Groups/Decision-RiskMGT/Risk/Citi Risk Modeling/Cards Corp/Projects/Custom Model Development/ECM/Alternative_Techniques/snapshot/MAR14_RAW_SAMPLE1.csv",header = TRUE)
indep= select(train,-FICO
,-segmentn)
indep2=select(indep,-bad_7wgfb_12)
depen=select(indep,bad_7wgfb_12)
indep$bad_7wgfb_12=factor(indep$bad_7wgfb_12)
rf_check = randomForest(bad_7wgfb_12~.,data=indep,mtry=14, ntree=3000,
replace=FALSE, classwt=NULL,
sampsize = 202745,
nodesize = 100,
maxnodes = NULL,
importance=TRUE, localImp=FALSE, nPerm=1,
norm.votes=TRUE, do.trace=100,
keep.forest=TRUE)