One-hot-encoding multi-byte string values in R-CodePudding

I collected some data from a survey that asked respondents to rank their preferences for players' profiles:

profile1: Tom, center, pitcher
profile2: Pete, right, hitter
profile3: Clay, left, hitter
profile4: Tom, right, fielder
profile5: Pete, left, fielder
profile6: Clay, center, pitcher

However, being unfamiliar with this questionnaire development software, the responses I collected are stored as multi-byte string values like the following (for each respondent), which are then read into R:

preferences <- data.frame(pref = c("1. Pete, right, hitter\n2. Clay, center, pitcher\n3. Tom, right, fielder\n4. Tom, center, pitcher\n5. Clay, left, hitter\n6. Pete, left, fielder",
"1. Tom, right, fielder\n2. Clay, center, pitcher\n3. Pete, left, fielder\n4. Pete, right, hitter\n5. Tom, center, pitcher\n6. Clay, left, hitter",
"1. Clay, left, hitter\n2. Tom, center, pitcher\n3. Pete, right, hitter\n4. Pete, left, fielder\n5. Clay, center, pitcher\n6. Tom, right, fielder"))

I'm wondering if there is any way to map each of a respondent's ranked choices to distinct column values corresponding to players' profiles given above, kind of like one-hot-encoding (OHE), and turn the result into the following format:

df <- data.frame(profile1 = c(4, 5, 2), profile2 = c(1, 4, 3), profile3 = c(5, 6, 1), profile4 = c(3, 1, 6), profile5 = c(6, 3, 4), profile6 = c(2, 2, 5))

df

  profile1 profile2 profile3 profile4 profile5 profile6
1        4        1        5        3        6        2
2        5        4        6        1        3        2
3        2        3        1        6        4        5

Any suggestions would be appreciated.

CodePudding user response：

preferences <- data.frame(pref = c("1. Pete, right, hitter\n2. Clay, center, pitcher\n3. Tom, right, fielder\n4. Tom, center, pitcher\n5. Clay, left, hitter\n6. Pete, left, fielder",
"1. Tom, right, fielder\n2. Clay, center, pitcher\n3. Pete, left, fielder\n4. Pete, right, hitter\n5. Tom, center, pitcher\n6. Clay, left, hitter",
"1. Clay, left, hitter\n2. Tom, center, pitcher\n3. Pete, right, hitter\n4. Pete, left, fielder\n5. Clay, center, pitcher\n6. Tom, right, fielder"), stringsAsFactors = F)

profiles <- c(
  "Tom, center, pitcher",
  "Pete, right, hitter",
  "Clay, left, hitter",
  "Tom, right, fielder",
  "Pete, left, fielder",
  "Clay, center, pitcher"
)


df <- data.frame(do.call(rbind, lapply(preferences$pref, function(x) {
  match(
   profiles,
   str_replace_all(strsplit(x, "\\n")[[1]], "^[0-9] . ", "")
  )
})))

names(df) <- paste0("profile", 1:length(profiles))

df

#   profile1 profile2 profile3 profile4 profile5 profile6
# 1        4        1        5        3        6        2
# 2        5        4        6        1        3        2
# 3        2        3        1        6        4        5

CodePudding user response：

You can create a lookup table with the profiles (lookup) and then manipulate the preferences object like this:

# Create data frame with six columns using `strsplit`
df=setNames(as.data.frame(tstrsplit(preferences$pref, "\\n")), paste0("profile",1:6))

# pivot longer and merge with lookup, then pivot back to wide
df %>% mutate(id = row_number()) %>% 
  pivot_longer(starts_with("profile"),names_prefix = "profile") %>% 
  mutate(value = str_remove(value,"^\\d [.] ")) %>% 
  inner_join(lookup, by=c("value" = "text")) %>% 
  pivot_wider(id_cols = id, names_from=profile, values_from = name,names_sort = TRUE,names_prefix = "profile") %>% 
  select(-id)

Output:

  profile1 profile2 profile3 profile4 profile5 profile6
  <chr>    <chr>    <chr>    <chr>    <chr>    <chr>   
1 4        1        5        3        6        2       
2 5        4        6        1        3        2       
3 2        3        1        6        4        5

Input (lookup table)

structure(list(profile = c("1", "2", "3", "4", "5", "6"), text = c("Tom, center, pitcher", 
"Pete, right, hitter", "Clay, left, hitter", "Tom, right, fielder", 
"Pete, left, fielder", "Clay, center, pitcher")), row.names = c(NA, 
-6L), class = "data.frame")

The lookup table appears like this:

  profile                  text
1       1  Tom, center, pitcher
2       2   Pete, right, hitter
3       3    Clay, left, hitter
4       4   Tom, right, fielder
5       5   Pete, left, fielder
6       6 Clay, center, pitcher

CodePudding user response：

In Base R you will do:

First reading your profiles:

text <- "profile1: Tom, center, pitcher
profile2: Pete, right, hitter
profile3: Clay, left, hitter
profile4: Tom, right, fielder
profile5: Pete, left, fielder
profile6: Clay, center, pitcher"

a <- read.dcf(textConnection(text), all = TRUE)

Note that if your profiles are in a file, then use a <- read.dcf('file.name', all = TRUE)

b <- strsplit(gsub("\\d ..", '', preferences$pref), '\n') 
setNames(data.frame(t(mapply(match, list(a), b))), names(a))

 profile1 profile2 profile3 profile4 profile5 profile6
1        4        1        5        3        6        2
2        5        4        6        1        3        2
3        2        3        1        6        4        5