dat <- dna_sj_n
dat$X1 <- as.numeric(dat$X1)

# Make split index
train_index <- sample(1:nrow(dat), nrow(dat)*0.75)

# Full data set
data_variables <- as.matrix(dat[,-1])
data_label <- dat[,"X1"]

data = as.matrix(dat)
data_matrix <- xgb.DMatrix(data, label = dat$X1)


# split train data and make xgb.DMatrix

train_data   <- data_variables[train_index,]
train_label <- data_label[train_index,]
train_label$X1 <- as.numeric(train_label$X1)

train_dat <- dat[train_index,]
test_dat <- dat[-train_index,]

train_matrix <- xgb.DMatrix(matrix(as.numeric(train_data),dim(train_data)), label = train_label$X1)

# split test data and make xgb.DMatrix
test_data  <- data_variables[-train_index,]
test_label <- data_label[-train_index,]
test_label$X1 <- as.numeric(test_label$X1)

test_matrix <- xgb.DMatrix(matrix(as.numeric(test_data),dim(test_data)), label = test_label$X1)



numberOfClasses <- length(unique(dat$X1))


xgb_params <- list("objective" = "multi:softprob","eval_metric" = "rmse", "num_class" = numberOfClasses)
nr    <- 55 # number of XGBoost rounds
cv.nfold  <- 10

# Fit cv.nfold * cv.nround XGB models and save OOF predictions
ptm <- proc.time()
ptm <- proc.time()
cvm0 <- xgb.cv(params = xgb_params,data = train_matrix, nrounds = nr, nfold = cv.nfold,verbose = FALSE,prediction = TRUE)
proc.time() - ptm

p0 <- data.frame(cvm0$pred) %>%   mutate(max_prob = max.col(., ties.method = "last"),        label = train_label + 1)
confusionMatrix(table(as.vector(as.matrix(p0$label[1])), as.vector(as.matrix(p0$max_prob))))

#########################################

xgb_params <- list("objective" = "multi:softprob","eval_metric" = "mlogloss",  "num_class" = numberOfClasses)
nr    <- 55 # number of XGBoost rounds
cv.nfold  <- 10

# Fit cv.nfold * cv.nround XGB models and save OOF predictions

ptm <- proc.time()
#cvm_best <- xgb.cv(params = xgb_params,data = train_matrix, nrounds = nr, nfold = cv.nfold,verbose = FALSE,prediction = TRUE,
#               max_depth = 6, eta = 0.2, min_child_weight=5.47,subsample=0.608,  colsample_bytree=1,
#               gamma=0,seed=500)

cvm_best <- xgb.cv(params = xgb_params,data = train_matrix, nrounds = nr, nfold = cv.nfold,verbose = FALSE,prediction = TRUE,
               max_depth = 6, eta = 0.2, min_child_weight=5,subsample=0.6,  colsample_bytree=1,
               gamma=0,seed=88)

proc.time() - ptm

p_best <- data.frame(cvm_best$pred) %>%   mutate(max_prob = max.col(., ties.method = "last"),        label = train_label + 1)
confusionMatrix(table(as.vector(as.matrix(p_best$label[1])), as.vector(as.matrix(p_best$max_prob))))