boost performance

expedia.boost.r

library(data.table)

#set sample size of test data
smp_size <- 10000
## set the seed to make your partition reproductible
set.seed(1234)
idx <- sample(seq_len(nrow(expedia_train)), size = smp_size)
test <- expedia_train[idx,]
train<-expedia_train[-idx,]

# how many test is going to do.
num_of_test <- 40
# by interval
by_val<-.5
# result container
final <- numeric(num_of_test)
# counter
k<-1 
# sequence of test value
run_times<-seq(0,round(num_of_test*by_val), by=by_val)

for (i in run_times) {
  print(i)
  prop <- i * .01
  sum_and_count <- function(x) {
    sum(x) * (1 - prop) + length(x) * prop
  }
  top_n <- function(hc, v1,minVal = 5) {
    hc_sorted <- hc[order(v1, decreasing = T)]
    n <- min(minVal, length(hc_sorted))
    paste(hc_sorted[1:n], collapse = "|")
  }
  
  hotel_with_score <-
    train[, sum_and_count(is_booking),by = list(hotel_market,hotel_cluster)]
  
  dest_top_n <-
    hotel_with_score[,top_n(hotel_cluster, V1), by = hotel_market]
  
  # generate result
  output <- merge(test,dest_top_n,by = "hotel_market")[,list(V1)]
  #table(output)
  setnames(output, 'hotel_cluster')
  
  # test the result
  result <- logical(length(output$hotel_cluster))
  for (j in 1:length(output$hotel_cluster)) {
    patt <- output$hotel_cluster[j]
    result[j] <- grepl(pattern = patt, x = test$hotel_cluster[j])
  }
  
  # store it in final to say how many true we have
  final[k] <- prop.table(table(result))['TRUE']
  print(final[k])
  k<-k+1
}
# which value produce the best outcome
run_times[which(final == max(final))]
# the maximum MAP@5
max(final)

# plot graph
plot(y = final, x = run_times)
lines(final)