Apply vs For loop in R

问题:

I wrote the following code to scrap tendering information from a portal on daily basis.

packages <- c('rvest', 'stringi', 'tidyverse','lubridate','dplyr')
purrr::walk(packages, library, character.only = TRUE, warn.conflicts = FALSE)
start_time <- proc.time()

Main Page to scrap and get total no of records.

data <- read_html('https://eprocure.gov.in/mmp/latestactivetenders')
total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
All_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
links_fair <- html_attr(links,'href')
links_fair <- links_fair[grep("tendersfullview",links_fair)]
All_tenders <- cbind(All_tenders,links_fair)

Reading the total number of records to fetch

Count_of_Recs_raw <- html_nodes(data, xpath = '//*[(@id = "edit-l-active-teners")]//div')
Count_of_Recs <- as.numeric(gsub("Total Tenders : ","",html_text(Count_of_Recs_raw[1])))

Functions for cleaning and processing data fields like dates and Factors.

process_dates <- function(data){
    cols2date <- c('Bid.Submission.Closing.Date','epublished_date','document_download_start_date','bid_submission_start_date','bid_opening_date','document_download_end_date','bid_submission_end_date')
    date_processed_data <- data
    date_processed_data[cols2date] <- lapply(data[cols2date] , dmy_hm)
    return(date_processed_data)
}

clean_process_data <- function(data){
    cols2factor <- c('State.Name','product_category','pre_qualification','organisation_name','organisation_type','tender_type')
    clean_processed_data <- data
    clean_processed_data[cols2factor] <- lapply(data[cols2factor] , factor)
   #clean_processed_data <- process_dates(clean_processed_data)
    return(clean_processed_data)

}

The code below is where precisely my question lies...

Table Scrapping starts here. Page one has already been scrapped to get the structure of the data frame.

for (page_no in 2:round(Count_of_Recs/10)){
    closeAllConnections()
    on.exit(closeAllConnections())
    url_bit1 <- 'https://eprocure.gov.in/mmp/latestactivetenders/page='
    url <- paste(url_bit1, page_no, sep="")
    cat(page_no,"\t",proc.time() - start_time,"\n")
    data <- read_html(url)
    total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
    Page_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
    links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
    links_fair <- html_attr(links,'href')
    links_fair <- links_fair[grep("tendersfullview",links_fair)]
    Page_tenders <- cbind(Page_tenders,links_fair)
    All_tenders <- rbind(All_tenders,Page_tenders)
 }

This for loop usually ends up taking hours to complete. I am looking for using the apply family to good effect so as to save on time. This program has further responsibility of fetching and processing all records and then for each individual record again scrapping a entirely new page every time (code not listed here)....

I have tried the following code but it doesn't give me what i want:

url_bit1 <- 'https://eprocure.gov.in/mmp/latestactivetenders/page='
read_page <- function(datain){
   closeAllConnections()
   on.exit(closeAllConnections())
   url <- paste(url_bit1, datain$S.No., sep="")
   cat(S.No.,"\t",proc.time() - start_time,"\n")
   data <- read_html(url)
   total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
   Page_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
   links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
   links_fair <- html_attr(links,'href')
   links_fair <- links_fair[grep("tendersfullview",links_fair)]
   Page_tenders <- cbind(Page_tenders,links_fair)
   All_tenders <- rbind(All_tenders,Page_tenders)
}

All_tenders <- sapply(All_tenders, FUN=read_page(All_tenders$S.No.))

Any advise, guidance, suggestions, inputs or help is welcome. I have been using R for 3-4 months only. I am also aware of Python's strengths in this matter over R but am inclined towards R for the solution to this problem.

回答1:

Your sapply function is incorrect. I made some edit on your code and tested it on sample size N = 50. We might use system.time() to find out how much time it takes to finish the task.

The "for" approach:

system.time(
  for (page_no in 1:50){
    closeAllConnections()
    on.exit(closeAllConnections())
    url_bit1 <- 'https://eprocure.gov.in/mmp/latestactivetenders/page='
    url <- paste(url_bit1, page_no, sep="")
    cat(page_no,"\t",proc.time() - start_time,"\n")
    data <- read_html(url)
    total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
    Page_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
    links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
    links_fair <- html_attr(links,'href')
    links_fair <- links_fair[grep("tendersfullview",links_fair)]
    Page_tenders <- cbind(Page_tenders,links_fair)
    All_tenders <- rbind(All_tenders,Page_tenders)
  }
)

#user  system elapsed 
# 50.15   81.26  132.73

The "lapply" approach:

All_tenders = NULL
url_bit1 <- 'https://eprocure.gov.in/mmp/latestactivetenders/page='
read_page <- function(datain){
  closeAllConnections()
  on.exit(closeAllConnections())
  url <- paste(url_bit1, datain, sep="")
  cat(datain,"\t",proc.time() - start_time,"\n")
  data <- read_html(url)
  total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
  Page_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
  links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
  links_fair <- html_attr(links,'href')
  links_fair <- links_fair[grep("tendersfullview",links_fair)]
  Page_tenders <- cbind(Page_tenders,links_fair)
  All_tenders <- rbind(All_tenders,Page_tenders)
}

system.time(
  All_tenders <- lapply(1:50, function(x) read_page(x))
)
# user  system elapsed 
# 49.84   78.97  131.16

If we want to put our results in a dataframe, then transform All_tenders list to a dataframe as follows:

All_tenders = do.call(rbind, lapply(All_tenders, data.frame, stringsAsFactors=FALSE)

Turns out lapply is slightly faster.

回答2:

for loops and sapply works differently: - for loops do stuff iteratively: they do computation on first element, then on second... - sapply do stuff on the list of elements independently (and in any order). So resulsts are constructed indepedently.

So at the and of your for loop, when you do:

All_tenders <- rbind(All_tenders,Page_tenders)

All_tenders variable increase itertively.

While in your sapply function, it won't work (since it doesn't know the results for other elements).

So you should do something like that:

url_bit1 <- 'https://eprocure.gov.in/mmp/latestactivetenders/page='
read_page <- function(datain){
   closeAllConnections()
   on.exit(closeAllConnections())
   url <- paste(url_bit1,  datain, sep="")
   cat(S.No.,"\t",proc.time() - start_time,"\n")
   data <- read_html(url)
   total_tenders_raw <- html_nodes(data,xpath = '//*[(@id = "table")]')
   Page_tenders <- data.frame(html_table(total_tenders_raw, header = TRUE))
   links <- html_nodes(data, xpath='//*[(@id = "table")] | //td | //a')
   links_fair <- html_attr(links,'href')
   links_fair <- links_fair[grep("tendersfullview",links_fair)]
   Page_tenders <- cbind(Page_tenders,links_fair)
   return(Page_tenders)
}

To return a result for each page and apply it the following way:

All_tenders_tmp <- sapply(2:round(Count_of_Recs/10), FUN=read_page)

Then your result will be list of all results and you can merge it with data.table::rbindlist for example.

I hope i was clear.