brunch_r_xgboost

gistfile1.txt

## apply xgboost on otto data

## url : https://www.kaggle.com/c/otto-group-product-classification-challenge/data

## reference : https://www.analyticsvidhya.com/blog/2016/01/xgboost-algorithm-easy-steps/



install.packages("Matrix")

library(Matrix)



install.packages(c("caret", "car", "dplyr"))

library(xgboost)

library(readr)

library(stringr)

library(caret)

library(car)

library(dplyr)



setwd("C:\\Users\\ts93856\\Desktop\\datasource")

# load data

df_train <- read.csv("train.csv")

df_test <- read.csv("test.csv")



df_test <- lapply(df_test, as.numeric)

df_test <- as.data.frame(df_test)

x_test <- df_test[, -1]



## very simple way to convert categorical data into numeric data

## xgboost는 numeric만 처리할 수 있음. 명목변수 -> 연속형 변수로 변형이 필요

x <- lapply(df_train, as.numeric)

train <- as.data.frame(x)

train <- train[, -ncol(train)]

train <- train[, -1]



## made target data

y <- x$target

y <- y-1



unique(y)



?xgboost

View(data.matrix(train))

## modeling

xgb <- xgboost(data = data.matrix(train), 

               label = y, 

               eta = 0.3, ## eta 학습률, x의 움직임 (default = 0.5)

               max_depth = 15,  ## max_depth, decision tree가 몇번 들어가는지

               nround=25, ## nround 최대로 iteration 돌아가는 횟수 

               subsample = 1,

               colsample_bytree = 0.5,

               seed = 1,

               eval_metric = "merror", ## 랜덤포레스트의 평가 지표

               objective = "multi:softprob",

               num_class = 9,

               nthread = 3

)



xgb



## scoring

y_pred <- predict(xgb, data.matrix(x_test), na.action = na.pass)

sum(y_pred)



## prediction

test_prediction <- matrix(y_pred, nrow = 9,

                          ncol=length(y_pred)/9) %>%

  t() %>%

  data.frame() %>%

  mutate(label = 1,

         max_prob = max.col(., "last"))



head(test_prediction, 3)

result <- test_prediction$max_prob



## submission file 작성

sub_csv <- matrix(0, nrow = nrow(x) , ncol = 9)

x <- data.frame(sample_sub , result)



for (i in 1:nrow(x)) {

  if (x$result[i] == 1){

    sub_csv[i,] <- c(1,0,0,0,0,0,0,0,0)

  } else if (x$result[i] == 2){

    sub_csv[i,] <- c(0,1,0,0,0,0,0,0,0)

  } else if (x$result[i] == 3){

    sub_csv[i,] <- c(0,0,1,0,0,0,0,0,0)

  } else if (x$result[i] == 4){

    sub_csv[i,] <- c(0,0,0,1,0,0,0,0,0)

  } else if (x$result[i] == 5){

    sub_csv[i,] <- c(0,0,0,0,1,0,0,0,0)

  } else if (x$result[i] == 6){

    sub_csv[i,] <- c(0,0,0,0,0,1,0,0,0)

  } else if (x$result[i] == 7){

    sub_csv[i,] <- c(0,0,0,0,0,0,1,0,0)

  } else if (x$result[i] == 8){

    sub_csv[i,] <- c(0,0,0,0,0,0,0,1,0)

  } else {

    sub_csv[i,] <- c(0,0,0,0,0,0,0,0,1)

  } 

}      



id <- seq(1:nrow(x))

result_submission <- data.frame(id, sub_csv)



write.csv(result_submission,"sampleSubmission.csv")

colnames(result_submission) <- c("id", "Class_1","Class_2",

                                 "Class_3",

                                 "Class_4",

                                 "Class_5",

                                 "Class_6",

                                 "Class_7",

                                 "Class_8",

                                 "Class_9")

nrow(result_submission)

colnames(result_submission)







## 가장 중요한 변수가 무엇인지 시각화하여 확인

# Lets start with finding what the actual tree looks like

model <- xgb.dump(xgb, with.stats = T)

model[1:10] #This statement prints top 10 nodes of the model

# Get the feature real names

names <- dimnames(data.matrix(train))[[2]]

# Compute feature importance matrix

importance_matrix <- xgb.importance(names, model = xgb)

# Nice graph

xgb.plot.importance(importance_matrix[1:10,], top_n = 10)







## 검정 과정

# pearson's validation

test <- chisq.test(train$feat_11, data)

print(test)