get_catalog_mapdcpsc <-
function( data_name = "mapdcpsc" , output_dir , ... ){
catalog <- NULL
for( cpsc_url in c( "https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MCRAdvPartDEnrolData/Monthly-Enrollment-by-Contract-Plan-State-County?items_per_page=100&items_per_page_options%5B5%5D=5%20per%20page&items_per_page_options%5B10%5D=10%20per%20page&items_per_page_options%5B25%5D=25%20per%20page&items_per_page_options%5B50%5D=50%20per%20page&items_per_page_options%5B100%5D=100%20per%20page&combine=&page=0" , "https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MCRAdvPartDEnrolData/Monthly-Enrollment-by-Contract-Plan-State-County?items_per_page=100&items_per_page_options%5B5%5D=5%20per%20page&items_per_page_options%5B10%5D=10%20per%20page&items_per_page_options%5B25%5D=25%20per%20page&items_per_page_options%5B50%5D=50%20per%20page&items_per_page_options%5B100%5D=100%20per%20page&combine=&page=1" ) ){
all_dates <- rvest::html_table(xml2::read_html(cpsc_url))
all_dates <- data.frame( all_dates[[1]] )[ , "Report.Period"]
all_links <- rvest::html_nodes(xml2::read_html(cpsc_url),xpath='//td/a')
prefix <- "https://www.cms.gov/"
these_links <-
gsub( "\">(.*)" , "" ,
gsub( '<a href=\"' , prefix ,
all_links ) )
this_catalog <-
data.frame(
output_filename = paste0( output_dir , "/" , all_dates , " cpsc enrollment.rds" ) ,
full_url = as.character( these_links ) ,
year_month = all_dates ,
stringsAsFactors = FALSE
)
for( this_row in seq( nrow( this_catalog ) ) ){
link_text <- readLines( this_catalog[ this_row , 'full_url' ] )
link_line <- grep( "cpsc(.*)zip|zip(.*)cpsc|onthly(.*)zip|nrollment(.*)zip" , link_text , value = TRUE )
link_line <- gsub( '(.*) href=(\"|\')' , "" , gsub( '(.*) href=\"/' , prefix , link_line ) )
this_catalog[ this_row , 'full_url' ] <- stringr::str_trim( gsub( '(\"|\')>(.*)' , "" , link_line ) )
}
catalog <- rbind( catalog , this_catalog )
}
catalog <- catalog[ !grepl( 'contract' , basename( catalog[ , 'full_url' ] ) ) , ]
catalog[ order( catalog$year_month ) , ]
}
lodown_mapdcpsc <-
function( data_name = "mapdcpsc" , catalog , ... ){
on.exit( print( catalog ) )
tf <- tempfile()
for ( i in seq_len( nrow( catalog ) ) ){
# download the file
cachaca( catalog[ i , "full_url" ] , tf , mode = 'wb' )
# extract the contents of the zipped file
# into the current year-month-specific directory
# and (at the same time) create an object called
# `unzipped_files` that contains the paths on
# your local computer to each of the unzipped files
unzipped_files <- unzip_warn_fail( tf , exdir = np_dirname( catalog[ i , 'output_filename' ] ) )
this_cont <- data.frame( readr::read_csv( grep( "Contract_Info" , unzipped_files , value = TRUE ) , guess_max = 100000 ) )
this_enr <- data.frame( readr::read_csv( grep( "Enrollment_Info" , unzipped_files , value = TRUE ) , guess_max = 100000 ) )
names( this_cont ) <- gsub( "Contract.Number" , "Contract.ID" , names( this_cont ) , fixed = TRUE )
names( this_enr ) <- gsub( "Contract.Number" , "Contract.ID" , names( this_enr ) , fixed = TRUE )
names( this_enr ) <- gsub( "SSA.Code" , "SSA.State.County.Code" , names( this_enr ) , fixed = TRUE )
names( this_enr ) <- gsub( "FIPS.Code" , "FIPS.State.County.Code" , names( this_enr ) , fixed = TRUE )
this_cont$year_month <- catalog[ i , 'year_month' ]
this_enr$year_month <- catalog[ i , 'year_month' ]
this_enr$enrolled <- as.numeric( gsub( "," , "" , this_enr$Enrollment ) )
this_enr$Plan.ID <- as.numeric( this_enr$Plan.ID )
this_cont$Plan.ID <- as.numeric( this_cont$Plan.ID )
this_enr <- this_enr[ , c( 'year_month' , 'Contract.ID' , 'Plan.ID' , 'FIPS.State.County.Code' , 'enrolled' ) ]
this_enr <- unique( this_enr )
this_cont <- unique( this_cont )
x <- merge( this_cont , this_enr )
stopifnot( nrow( x ) == nrow( this_enr ) )
# convert all column names to lowercase
names( x ) <- tolower( names( x ) )
x$year_month <- catalog[ i , 'year_month' ]
names( x ) <- gsub( "\\." , "_" , names( x ) )
names( x )[ names( x ) == 'fips_state_county_code' ] <- 'fips'
saveRDS( x , file = catalog[ i , 'output_filename' ] , compress = FALSE )
# add the number of records to the catalog
catalog[ i , 'case_count' ] <- nrow( x )
# delete the temporary files
file.remove( tf , unzipped_files )
cat( paste0( data_name , " catalog entry " , i , " of " , nrow( catalog ) , " stored at '" , catalog[ i , 'output_filename' ] , "'\r\n\n" ) )
}
on.exit()
catalog
}
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.