Agfdhyk

I have data in a table in which one cell in every row is a multiline string, which is formatted a a bit like a document with references at the end of it. For example, one of those strings looks like:

item A...1

item B...2

item C...3

item D...2

1=foo

2=bar

3=baz

My eventual goal is to extract foo/bar/baz into columns and count the matching items. So for the above, I'd end up with a row including:

foo | bar | baz

----+-----+----

1   | 2   | 1

I tried to start by extracting the "reference" mappings, as a nested data.table looking like this:

code | reason

-----+-------

1    | foo

2    | bar

3    | baz

Here's how I tried to do it, using data.table and stringr.

encounter_alerts[, whys := lapply(

  str_extract_all(text, regex('^[0-9].*$', multiline = TRUE)),

  FUN = function (s) { fread(text = s, sep = '=', header = FALSE, col.names = c('code', 'reason')) }

)]

I am very confused by the error message I get when I try to do this:

Error in fread(text = s, sep = "=", header = FALSE, col.names = c("code",  :

  file not found: 1=foo

I am explicitly using text rather than file so I'm not sure how it's trying to interpret the line of text as a filename!

When I test this with a single row, it seems to work fine:

> fread(text = str_extract_all(encounter_alerts[989]$text, regex('^[0-9].*$', multiline = TRUE))[[1]], sep = '=', header = FALSE, col.names = c('code', 'reason'))

   code reason

1:    1    foo

2:    2    bar

What am I doing wrong? Is there a better way to do this?

Thanks!

Note: Edited after reading comments

From your comment, I tried to reproduce what I understand your data might look like.

library(tidyverse)



df <- tibble(

  strings = c("item A...1

item B...2

item C...3

item D...2

1=foo

2=bar

3=baz",

"item A...2

item B...2

item C...3

item D...1

1=toto

2=foo

3=lala",

"item A...3

item B...3

item C...3

item D...1

1=tutu

3=ttt")

)

Code:

get_ref <- function(string) {

  string %>%

    str_split("n") %>%

    unlist() %>% 

    str_subset("=") %>%

    str_split_fixed("=", 2) %>%

    as_tibble() %>%

    rename(code = V1, reason = V2)

}



list1 <- map(df$strings, get_ref)



get_value <- function(string) {

  string %>%

      str_split("n") %>%

      unlist() %>% 

      str_subset("\.\.\.") %>%

      str_replace_all(".*\.\.\.", "") %>%

      as_tibble() %>%

    rename(code = value)

}



list2 <- map(df$strings, get_value)



get_result <- function(df1, df2) {

  left_join(df1, df2) %>%

    count(reason) %>%

    spread(reason, n)

}



result <- map2_df(list1, list2, get_result)



result[is.na(result)] <- 0



result

Result

# A tibble: 3 x 7

    bar   baz   foo  lala  toto   ttt  tutu

  <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>

1     2     1     1     0     0     0     0

2     0     0     2     1     1     0     0

3     0     0     0     0     0     3     1

using stringr and dplyr you can do it easily

library(stringr)

library(dplyr)

v <- as.data.frame(c(  "item A...1",

         "item B...2",

        "item C...3",

         "item D...2"))

colnames(v)<- "items"



matching <- c( "1",

               "2",

               "3")

Mapping <- read.table(text="code     reason

1    foo

                      2  bar

                      3  baz

                      ", header = T)



## Answer 

df1<- v %>%

  mutate(code = str_extract(v$items, str_c(matching, collapse = "|")))

str(df1)

str(Mapping)

df1$code <- as.numeric(df1$code )



df1 <- left_join(df1,Mapping)

please have a look

There's probably a nicer way to do this, but here's a solution that doesn't require any additional libraries (beyond stringr, which you're already using).

sample_str <- 'item A...1

item B...2

item C...3

item D...2

1=foo

2=bar

3=baz'



lines <- stringr::str_split(sample_str, 'n', simplify = T)



extracted_strs <- lines[stringr::str_detect(lines, '^\d=\w+$')]



dfs_list <- lapply(extracted_strs, function(x) {

  str_parts <- stringr::str_split(x, '=', simplify = T)

  df_args = list()

  df_args[[str_parts[2]]] = as.integer(str_parts[1])

  df_args[['stringsAsFactors']] = F



  do.call(data.frame, df_args)

})





df <- do.call(cbind, dfs)

Thanks so much to @prosoitos for helping with this. Here's the final code I ended up using, highly based on the accepted answer — it's a mix of different packages and so forth which I hope to clean up eventually, but deadlines happen...

get_code_reason_mapping <- function(alert_text) {

  alert_text %>%

    str_extract_all(regex('^[0-9]=(.*)$', multiline = T)) %>%

    unlist() %>%

    str_split_fixed("=", 2) %>%

    as.data.table() %>%

    setnames(c('code', 'reason'))

}



encounter_alerts$code_reason_mapping <- map(encounter_alerts$alert_text, get_code_reason_mapping)



get_why_codes <- function(alert_text) {

  alert_text %>%

    str_extract_all(regex('[/n][0-9e][0-9>][0-9]$', multiline = TRUE)) %>%

    unlist() %>%

    str_sub(-1) %>%

    as.data.table() %>%

    setnames(c('code'))

}



encounter_alerts$why_codes <- map(encounter_alerts$alert_text, get_why_codes)



get_code_counts <- function(df1, df2) {

  left_join(df1, df2) %>%

    count(reason) %>%

    spread(reason, n)

}



code_counts <- map2_df(encounter_alerts$code_reason_mapping, encounter_alerts$why_codes, get_code_counts)



code_counts[is.na(code_counts)] <- 0



code_counts