Commit: 9dbc6c80a640c7f2c27cde80898f2942940db6e6 - myprojects-hg-reborn (hg)

Commit MetaInfo

• Revision: 9dbc6c80a640c7f2c27cde80898f2942940db6e6 (tree)
• Time: 2014-12-15 04:41:39
• Author: Lorenzo Isella <lorenzo.isella@gmai...>
• Commiter: Lorenzo Isella

Log Message

A code to use caret + randomForest. In particular, it allows to deal with unbalanced classes.

Change Summary

• add: R-codes/analyze-transactions.R

Incremental Difference

diff -r 5f7f832e74b3 -r 9dbc6c80a640 R-codes/analyze-transactions.R
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/R-codes/analyze-transactions.R	Sun Dec 14 20:41:39 2014 +0100
@@ -0,0 +1,249 @@
+rm(list=ls())
+
+#library(ada)
+## library(ggplot2)
+#library(glmnet)
+#library(reshape2)
+library(randomForest)
+## require(gridExtra)
+## library(corrplot)
+## library(scales)
+## library(digest)
+library(caret)
+## library(gbm)
+#library(doMC)
+## library(nnet)
+## library(gbm)
+
+
+
+confusion.glm <- function(data, model) {
+prediction <- ifelse(predict(model, data, type='response') > 0.5, TRUE, FALSE)
+confusion <- table(prediction, as.logical(model$y))
+confusion <- cbind(confusion, c(1 - confusion[1,1]/(confusion[1,1]+confusion[2,1]), 1 - confusion[2,2]/(confusion[2,2]+confusion[1,2])))
+confusion <- as.data.frame(confusion)
+names(confusion) <- c('FALSE', 'TRUE', 'class.error')
+confusion
+} 
+
+
+
+
+
+
+
+##########################################################################
+###########################################################################
+
+
+analyze_bin <- 1
+
+analyze_rev <- 1
+
+if (analyze_bin==1){
+
+set.seed(1234)
+
+
+    
+mydata <- readRDS("train_mf_bin.RDS")
+
+
+## see http://bit.ly/1uEEPze
+
+
+nmin <- sum(mydata$Sale_MF == 1)
+
+ ctrl <- trainControl(method = "cv",
+                      classProbs = TRUE,summaryFunction = twoClassSummary)
+
+
+set.seed(1234)
+
+
+print("RF for mf")
+
+rf_mf <- train(Sale_MF ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+                        metric = "ROC",
+                        trControl = ctrl,
+                        ## Tell randomForest to sample by strata. Here, 
+                        ## that means within each class
+                        strata = mydata$Sale_MF,
+                        do.trace=500,
+                        ## Now specify that the number of samples selected
+                        ## within each class should be the same
+                        sampsize = rep(nmin, 2))
+
+saveRDS(rf_mf, "rf_mf_bin.RDS")
+
+################################################
+
+
+mydata <- readRDS("train_cc_bin.RDS")
+
+nmin <- sum(mydata$Sale_CC == 1)
+
+print("RF for cc")
+
+
+rf_cc <- train(Sale_CC ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+                        metric = "ROC",
+                        trControl = ctrl,
+                        ## Tell randomForest to sample by strata. Here, 
+                        ## that means within each class
+                        strata = mydata$Sale_CC,
+                        do.trace=500,
+                        ## Now specify that the number of samples selected
+                        ## within each class should be the same
+                        sampsize = rep(nmin, 2))
+
+saveRDS(rf_mf, "rf_cc_bin.RDS")
+
+
+################################################
+
+
+mydata <- readRDS("train_cl_bin.RDS")
+
+nmin <- sum(mydata$Sale_CL == 1)
+
+print("RF for cl")
+
+
+rf_cl <- train(Sale_CL ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+                        metric = "ROC",
+                        trControl = ctrl,
+                        ## Tell randomForest to sample by strata. Here, 
+                        ## that means within each class
+                        strata = mydata$Sale_CL,
+                        do.trace=500,
+                        ## Now specify that the number of samples selected
+                        ## within each class should be the same
+                        sampsize = rep(nmin, 2))
+
+saveRDS(rf_cl, "rf_cl_bin.RDS")
+
+
+
+
+
+#######################################################################
+#######################################################################
+#######################################################################
+#######################################################################
+
+
+
+}
+
+if (analyze_rev==1){
+
+set.seed(1234)
+
+
+ ctrl <- trainControl(method = "cv")
+
+    
+mydata <- readRDS("train_mf_amount.RDS")
+
+
+
+sel <- which(mydata$Revenue_MF > 0)
+
+mydata <- mydata[sel, ]
+
+
+
+
+
+
+print("RF for mf")
+
+rf_mf <- train(Revenue_MF ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+               ## metric = "ROC",
+               trControl = ctrl,
+                        do.trace=500)
+
+
+saveRDS(rf_mf, "rf_mf_amount.RDS")
+
+
+
+#####################################################################
+
+mydata <- readRDS("train_cc_amount.RDS")
+
+
+
+sel <- which(mydata$Revenue_CC > 0)
+
+mydata <- mydata[sel, ]
+
+
+ 
+
+print("RF for cc")
+
+rf_cc <- train(Revenue_CC ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+               ## metric = "ROC",
+               trControl = ctrl,
+                        do.trace=500)
+
+
+saveRDS(rf_cc, "rf_cc_amount.RDS")
+
+
+ 
+##################################################################
+
+
+mydata <- readRDS("train_cl_amount.RDS")
+
+
+
+sel <- which(mydata$Revenue_CL > 0)
+
+mydata <- mydata[sel, ]
+
+
+ 
+
+print("RF for cl")
+
+rf_cl <- train(Revenue_CL ~ ., data =mydata,
+                        method = "rf",
+                        ntree = 1500,
+                        tuneLength = 10,
+               ## metric = "ROC",
+               trControl = ctrl,
+                        do.trace=500)
+
+
+saveRDS(rf_cl, "rf_cl_amount.RDS")
+
+ 
+
+
+
+
+
+
+ 
+}
+
+print("So far so good")