Notes

library(caTools)
library(xgboost)

# creating a function to find best nrounds, using 70% data to train and 30% data to validate
validation_fun <- function(temp_train, temp_target)
{
targte_no. <- paste('targte',fun_call,sep='_')
cat("n",'training the model for',targte_no., 'to get the optimum number of iterations',"n")

#split the training for training and validation
set.seed(3000)

spl = sample.split(temp_target, SplitRatio = 0.70)
cv_train_data = subset(temp_train, spl == TRUE)
cv_train_target <- temp_target[spl==TRUE] # it will be used in xgb.DMatrix

cv_test_data = subset(temp_train, spl == FALSE)
cv_test_target <- temp_target[spl==FALSE] # it will be used in xgb.DMatrix


# summary of target variables
# summary(temp_target)
# summary(cv_train_target)
# summary(cv_test_target)

####### now the modelling part
dtrain<-xgb.DMatrix(data=data.matrix(cv_train_data),label=data.matrix(cv_train_target))
dval<-xgb.DMatrix(data=data.matrix(cv_test_data),label=data.matrix(cv_test_target))
watchlist<-list(val=dval)


param = list(objective ="reg:linear",
eval_metric = 'rmse',
booster = "gbtree",
"max_depth" = 10,
"eta" = 0.1,
"subsample" = 1,
"colsample_bytree" =1 ,
"min_child_weight" = 1)

set.seed(600)
bst = xgb.train(param=param,
data =dtrain,
nrounds=10000,
verbose = 1,
early_stopping_rounds = 50,
watchlist = watchlist ,
print_every_n = 20,
maximize = FALSE)

# saving the results to a file
cat ("n",fun_call, 'cv_best_nrounds:', bst$best_iteration,'nrounds:', as.integer(bst$best_iteration/0.7),
'cv_rmse_score:', bst$best_score,
file = "F:/Data Science/impulse/nrounds.txt", append = TRUE)


# printing to console
cat("n",'-------------------------------- CV Results -----------------------------------------')
cat ("n",'cv_best_nrounds:', bst$best_iteration,'nrounds:', as.integer(bst$best_iteration/0.7),
'cv_rmse_score:', bst$best_score )
cat("n",'-------------------------------------------------------------------------------------')

assign("final_nrounds",as.integer(bst$best_iteration/0.7),envir =.GlobalEnv)

cat("n",'Now training the actual model for',targte_no.,'which will be saved for later use',"n")

}