fit <- lm(Gross ~ Year + Runtime + Budget, data = movie)

predicted = predict(fit, movie)

rerror = rmse(predicted, movie$Gross)

rmse = function(predicted_vector, actual_vector) {
   difference = predicted_vector - actual_vector
   rmse = sqrt(mean(difference ^ 2))
   return(rmse)
}

rmse_frame = data.frame(size = numeric(20), train_rmse = numeric(20), test_rmse = numeric(20))
size_vector = seq(from = 5, to = 100, by = 5)

for(c in 1:20) {
  size = size_vector[c]
  total_rows_training = dim(training_set)[1]
  number_of_rows_to_select = ceiling(((size/100) *  total_rows_training))
  train_rmse_vector = c()
  test_rmse_vector = c()
  for(i in 1:10) { # RUN 10 TIMES, THEN TAKE AVERAGE OF RMSE
    # SHUFFLE THE TRAINING SET
    shuffle_training =  training_set[sample(nrow(training_set)),]
    # SELECT THE ROWS FOR TRAINING SET ACCORDING TO THE SAMPLE SIZE
    sample_train_set = shuffle_training[1:number_of_rows_to_select,]
    fit <- lm(Gross ~ Year + Runtime + Budget + imdbVotesTransformed , data = sample_train_set)
    train_prediction = predict(fit, sample_train_set)
    train_rmse = rmse(train_prediction, sample_train_set$Gross)
    train_rmse_vector = c(train_rmse_vector, train_rmse)
    # USE MODEL GENERATED ON THE TEST_SET
    test_prediction = predict(fit, test_set)
    test_rmse = rmse(test_prediction, test_set$Gross)
    test_rmse_vector = c(test_rmse_vector, test_rmse)
  }
  rmse_frame$size[c] = size
  # TAKE AVERAGE OF RMSE
  rmse_frame$train_rmse[c] = mean(train_rmse_vector)
  rmse_frame$test_rmse[c] = mean(test_rmse_vector)
}