问题
I have recently converted from ifelse to case_when from dplyr.
Aim
I would like to be able to detect more than one regex from a statement in a dataframe using case_when as follows:
Input
statement<-data.frame(statement = c("I have performed APC and RFA",
"An EMR was done","I didn't do anything"),stringsAsFactors=FALSE)
Desired output
statement out
I have performed APC and RFA APC,RFA
An EMR was done EMR
I didn't do anything No Event
Attempt
library(dplyr)
library(stringr)
dataframe <-
dataframe %>%
mutate(
EVENT = case_when(
str_detect(statement,"EMR") ~ "EMR",
str_detect(statement, "HALO|RFA") ~ "RFA",
str_detect(statement, "APC") ~ "APC",
TRUE ~ "No Event"
)
)
The problem
This gives only one output per statement rather than multiple output if multiple strings are present. Is there a way to detect multiple strings?
回答1:
1) gsubfn::strapply strapply can do the extraction and translation all at once. strapply will, for each component of stmt, match the pattern pat to it and all matches will be translated using L and then returned. The empty argument defines what is returned for components of stat having no matches. This gives a list of matches, one list component per row, on which toString is applied to convert each to a comma separated character string. This is the shortest of the 3 alternatives presented here.
library(gsubfn)
L <- list(APC = "APC", EMR = "EMR", HALO = "RFA", RFA = "RFA")
pat <- paste(names(L), collapse = "|")
transform(statement,
out = sapply(strapply(stmt, pat, L, empty = "No Event"), toString),
stringsAsFactors = FALSE)
giving:
stmt out
1 I have performed APC and RFA APC, RFA
2 An EMR was done EMR
3 I didn't do anything No Event
2) Base R Using L and pat from above, create a function which takes a character vector of words x and extracts out those words matched by pat into g. If g has non-zero length translate its elements using L and compress it into a single string using toString; otherwise, return No Event.
Now split each element of stmt into words using strsplit and apply process to each such character vector.
process <- function(x) {
g <- grep(pat, x, value = TRUE)
if (length(g)) toString(L[g]) else "No Event"
}
transform(statement, out = sapply(strsplit(stmt, "\\s+"), process),
stringsAsFactors = FALSE)
3) dplyr/tidyr Using L from (1) group by row number and stmt and separate the words into separate rows. Filter out those words in names(L) and collapse all rows in one stmt group translating through L and using toString to generate a comma separated string. Drop the n column. At this point we have the desired result except that No Event rows are still missing so right join what we have with statement and replace NAs with No Event.
library(dplyr)
library(tidyr)
statement %>%
group_by(n = 1:n(), out = stmt) %>%
separate_rows(out) %>%
filter(out %in% names(L)) %>%
summarize(stmt = stmt[1], out = toString(L[out])) %>%
ungroup %>%
select(-n) %>%
right_join(statement, by = "stmt") %>%
mutate(out = if_else(is.na(out), "No Event", out))
giving:
# A tibble: 3 x 2
stmt out
<chr> <chr>
1 I have performed APC and RFA APC, RFA
2 An EMR was done EMR
3 I didn't do anything No Event
Note
We used this as the input:
statement <- structure(list(stmt = c("I have performed APC and RFA",
"An EMR was done", "I didn't do anything")),
class = "data.frame", row.names = c(NA, -3L))
Updates
Have revised a number of times after re-reading the question. Also added more alternatives.
回答2:
The logic of case_when is that it doesn't execute the remaining conditions once a condition is met, so you can't actually get two outputs from a case_when statement. So if you want to use case_when it is advised to start with a least common condition and then slowly keep on making it more general. (hence, TRUE is the last condition)
If you want to stick with case_when you can add an additional condition and check for both the cases separately and give output accordingly.
library(dplyr)
statement %>%
mutate(
EVENT = case_when(
str_detect(statement, "APC") & str_detect(statement, "RFA") ~ "APC,RFA",
str_detect(statement,"EMR") ~ "EMR",
str_detect(statement, "HALO|RFA") ~ "RFA",
str_detect(statement, "APC") ~ "APC",
TRUE ~ "No Event"
)
)
# statement EVENT
#1 I have performed APC and RFA APC,RFA
#2 An EMR was done EMR
#3 I didn't do anything No Event
#4 FALSE No Event
回答3:
An idea via base R is to extract the words with all upper case letters, and paste the ones that are bigger than 1 character, i.e.,
sapply(regmatches(statement$statement, gregexpr('\\b[A-Z]+\\b', statement$statement)),
function(i) {
v1 <- i[nchar(i) > 1];
toString(v1)
})
#[1] "APC, RFA" "EMR" ""
回答4:
I don't think case_when is the best way to go. I think it depends a little bit on how many mappings like "HALO|RFA" you have. If there are many, it might be worth the time to write a proper function. However, if it is just this one, it might be faster to put together a pipe using dplyr verbs.
I would suggest using str_extract_all and unnest to get a tidy data frame with the relevant verbs and then use str_replace_all to resolve the mapping. In the end, I would use unique to make sure that we have no duplicate rows from the replacements.
Notice that the first column with APC, RFA will be split in two. I realize that this is not what you asked but it will make susequent processing in the tidyverse much easier. See this link for more on tidy data conventions: https://tidyr.tidyverse.org/articles/tidy-data.html
In it's current implementation, unnest will drop the last row where no pattern was matched. If you would like a NA instead, you can perform a full join with the original data. See also https://github.com/tidyverse/tidyr/issues/358
statement<-data.frame(statement = c("I have performed APC and RFA","An EMR was done","I didn't do anything"),stringsAsFactors=FALSE)
library(tidyverse)
statement %>% mutate(
out = statement %>%
str_extract_all("((EMR)|(HALO)|(RFA)|(APC))")
) %>% unnest(.drop =FALSE) %>%
mutate(
out = out %>% str_replace_all("HALO", "RFA")
) %>%
unique() %>%
full_join(statement)
The output will be
statement out
I have performed APC and RFA APC
I have performed APC and RFA RFA
An EMR was done EMR
I didn't do anything <NA>
来源：https://stackoverflow.com/questions/53851627/how-to-detect-more-than-one-regex-in-a-case-when-statement