At Santander our mission is to help people and businesses prosper. We are always looking for ways to help our customers understand their financial health and identify which products and services might help them achieve their monetary goals.
Our data science team is continually challenging our machine learning algorithms, working with the global data science community to make sure we can more accurately identify new ways to solve our most common challenge, binary classification problems such as: is a customer satisfied? Will a customer buy this product? Can a customer pay this loan?
In this challenge, we invite Kagglers to help us identify which customers will make a specific transaction in the future, irrespective of the amount of money transacted. The data provided for this competition has the same structure as the real data we have available to solve this problem.
Step#1 Load both train & test sets.
library(dplyr)
library(mlbench)
library(mxnet)
setwd("G:/DataScienceProject/Kaggle_santander-customer-transaction-prediction")
Train <- read.csv(file = "train.csv", header = TRUE)
Test <- read.csv(file = "test.csv", header = TRUE)
Train$ID_code <- as.character(Train$ID_code)
Test$ID_code <- as.character(Test$ID_code)
Test$target <- as.integer(0)
Test <- Test[,c(1,202,2:201)]
VarList <- read.csv(file = "VarList0.csv", header = TRUE)Step#2: Since we have more than 200 columns, it would be nice to reduce them into a small subset. The VarList (see above) is laready contains such list, as in early phase I ran GBM for analyze the most important columns for this.
#Col alignment
VarList$var <- as.character(VarList$var)
VarList$var <- gsub("var_", "", VarList$var)
VarList$var <- as.integer(VarList$var) + 3
VarList <- VarList[order(VarList$var),]
Train1 <- Train[,c(1,2)]
Test1 <- Test[,c(1,2)]
for (j in 1:length(VarList$var)) {
D <- paste0("var_", VarList[j,1] - 3)
Train1[,D] <- Train[,VarList[j,1]]
Test1[,D] <- Test[,VarList[j,1]]
j <- j + 1
}
Train <- Train1
Test <- Test1Step#3: We are going to use MxNet hence we need to convert all double types into integer.
for (i in 3:length(Train)) {
Train[,i] <- as.integer(round(Train[,i]))
Test[,i] <- as.integer(round(Test[,i]))
i <- i + 1
}
cv <- Train
Train1 <- Train[Train$target == 1,]Step#4 Now, we will craete a simple dataframe for “playing” with MxNet trees & learning rate for getting most accurate prediction.
## hidden_node learning.rate
## 1 256 0.01
## 2 128 0.01
## 3 64 0.01
## 4 256 0.05
## 5 128 0.05
## 6 64 0.05Step#5 Lets run machine learming! My model has 3 layers, as followed: Layer 1 with 6 nodes –> Layer 2 with 2 nodes –> Layer 3 (output) for actual prediction
for (i in 1:6) {
Train1.y <- as.array(Train1[,2])
Train1.x <- as.matrix(Train1[,3:42])
cv.y <- as.array(cv[,2])
cv.x <- as.matrix(cv[,3:42])
Test.y <- as.array(cv[,2])
Test.x <- as.matrix(cv[,3:42])
model1 <- mx.mlp(Train1.x, Train1.y, hidden_node=mlConf[i,1], out_node=2, out_activation="softmax",
num.round=20, array.batch.size=1004, learning.rate=mlConf[i,2], momentum=0.9,
eval.metric=mx.metric.accuracy)
preds1 <- t(predict(model1, cv.x))
preds1 <- as.data.frame(preds1)
names(preds1) <- c(paste0("L1-",i, "0"),paste0("L1-",i, "1"))
preds1[,1] <- as.integer(preds1[,1] * 100)
preds1[,2] <- as.integer(preds1[,2] * 100)
cv <- cbind(cv, preds1)
preds1 <- t(predict(model1, Test.x))
preds1 <- as.data.frame(preds1)
names(preds1) <- c(paste0("L1-",i, "0"),paste0("L1-",i, "1"))
preds1[,1] <- as.integer(preds1[,1] * 100)
preds1[,2] <- as.integer(preds1[,2] * 100)
Test <- cbind(Test, preds1)
i <- i + 1
}
cv <- cv[,c(1,2,43:54)]
Test <- Test[,c(1,2,43:54)]
Train1 <- cv[cv$target == 1,]
for (i in 1:2) {
Train1.y <- as.array(Train1[,2])
Train1.x <- as.matrix(Train1[,3:14])
cv.y <- as.array(cv[,2])
cv.x <- as.matrix(cv[,3:14])
Test.y <- as.array(cv[,2])
Test.x <- as.matrix(cv[,3:14])
model2 <- mx.mlp(Train1.x, Train1.y, hidden_node=128, out_node=2, out_activation="softmax",
num.round=20, array.batch.size=1004, learning.rate=0.01, momentum=0.9,
eval.metric=mx.metric.accuracy)
preds2 <- t(predict(model2, cv.x))
preds2 <- as.data.frame(preds2)
names(preds2) <- c(paste0("L2-",i, "0"),paste0("L2-",i, "1"))
preds2[,1] <- as.integer(preds2[,1] * 100)
preds2[,2] <- as.integer(preds2[,2] * 100)
cv <- cbind(cv, preds2)
preds2 <- t(predict(model2, Test.x))
preds2 <- as.data.frame(preds2)
names(preds2) <- c(paste0("L2-",i, "0"),paste0("L2-",i, "1"))
preds2[,1] <- as.integer(preds2[,1] * 100)
preds2[,2] <- as.integer(preds2[,2] * 100)
Test <- cbind(Test, preds2)
i <- i + 1
}
cv <- cv[,c(1,2,15:18)]
Test <- Test[,c(1,2,15:18)]
Train1 <- cv[cv$target == 1,]
Train1.y <- as.array(Train1[,2])
Train1.x <- as.matrix(Train1[,3:6])
cv.y <- as.array(cv[,2])
cv.x <- as.matrix(cv[,3:6])
Test.y <- as.array(cv[,2])
Test.x <- as.matrix(cv[,3:6])
model3 <- mx.mlp(Train1.x, Train1.y, hidden_node=64, out_node=2, out_activation="softmax",
num.round=20, array.batch.size=1004, learning.rate=0.05, momentum=0.9,
eval.metric=mx.metric.accuracy)
preds3 <- t(predict(model3, Test.x))
preds3 <- as.data.frame(preds3)
names(preds3) <- c(paste0("L3-",i, "0"),paste0("L3-",i, "1"))
preds3[,1] <- as.integer(preds3[,1] * 100)
preds3[,2] <- as.integer(preds3[,2] * 100)
Test <- cbind(Test, preds3)
Test <- Test[,c(1,2,7:8)]
for (i in 1:length(Test$ID_code)) {
if(Test[i,4] > Test[i,3]){
Test[i,2] <- 1
} else{}
i <- i + 1
}
Test <- Test[,c(1,2)]
write.csv(Test, file = "mySubmission.csv", row.names = FALSE)The accuracy didn’t raised above 90% although using several techniques MxNet & GBM (in another model testing. But is was fun. Thank you Kaggle.