xgboostr.r

xgboostr.r

# Training an XGBoost in R - used in blog post:
# https://towardsdatascience.com/data-science-tutorials-training-an-xgboost-using-r-cf3c00b1425

library(dplyr)
library(xgboost) 
library(Metrics)
library(ggplot2)

# Load london bike csv
london_bike <- read.csv('./london_merged.csv')

# Using a function to split into train and test
train_test_split <- function(data, percentage) {
  
  data_with_row_id <- data %>% 
    mutate(id = row_number())
  
  set.seed(1234)
  training_data <- data_with_row_id %>%
    sample_frac(percentage)
  test_data <- anti_join(
    data_with_row_id,
    training_data,
    by='id'
  )
  
  training_data$id <- NULL
  test_data$id <- NULL
  
  return (list(training_data, test_data))
}

# Keeping 80% for the training set
training_data <- train_test_split(
  london_bike, 0.8
  )[[1]]
test_data <- train_test_split(
  london_bike, 0.8
  )[[2]]

# Subsetting only the features
X_train <- training_data[,c('t1','t2','hum',
                            'wind_speed','weather_code',
                            'is_holiday','is_weekend',
                            'season')]

X_test <- test_data[,c('t1','t2','hum',
                       'wind_speed','weather_code',
                       'is_holiday','is_weekend',
                       'season')]

# Defining target variable
y_train <- training_data$cnt
y_test <- test_data$cnt

# Fitting XGBoost Model
set.seed(1234)
xgb <- xgboost(data = as.matrix(X_train),
               label = y_train,
                nround = 10)

# Acessing hyperparameters
?xgboost

# Timing the execution
system.time(
  xgboost(data = as.matrix(X_train),
                       label = y_train,
                       nround = 10))

# Assessing performance
rmse(
  y_test, 
  predict(xgb, as.matrix(X_test))
  )

# Adding a new hyperparameter
set.seed(1234)
xgb_ext <- xgboost(data = as.matrix(X_train),
               label = y_train,
               nround = 50,
               max_depth=20)

# Plotting RMSE throughout iterations
ggplot(
  data = xgb_ext$evaluation_log,
  aes(x=iter, y=train_rmse)
) + geom_line(color='darkred') + ylab('RMSE') + xlab('Iteration Number') + theme_light() + geom_point()

# Assessing execution time
system.time(xgboost(data = as.matrix(X_train),
                    label = y_train,
                    nround = 50,
                    max_depth=20)
)


# Assessing performance
rmse(
  y_test, 
  predict(xgb_ext, as.matrix(X_test))
)