问题
Is there a more efficient way to count the most frequently appearing digit in a string? My R code below calls gsub() 10 times for each string; and I have gazillions of strings to process.
> txt = 'wow:011 test 234567, abc=8951111111111aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa'
> max(vapply(0:9, function(i) nchar(gsub(paste0('[^',i,']'), '', txt)), integer(1L)))
[1] 12
I don't care about the digit itself. I just want the count of the most frequent one.
I would prefer to use R's core packages, unless some external package offers a significant outperformance. I use x64 R version 3.4.1 (2017-06-30) on Windows 10.
UPDATE:
Here is the (apples-to-apples) performance comparison of excellent suggestions below.
> microbenchmark(
+ original = max(vapply(0:9, function(i) nchar(gsub(paste0('[^',i,']'), '', s)), integer(1L))),
+ strsplit = max(table(unlist(strsplit(gsub("\\D+", "", s), "")))),
+ gregexpr = max(vapply(0:9, function(d) sum(unlist(gregexpr(d, s)) > 0), integer(1L))),
+ stringi = max(vapply(0:9, function(x) stri_count_fixed(s, x), integer(1L))),
+ raw=max(vapply(0x30:0x39, function(x) sum(charToRaw(s)==x), integer(1L))),
+ tabulate = max(tabulate(as.integer(charToRaw(paste('a',s))))[48:57]),
+ times=1000L)
Unit: microseconds
expr min lq mean median uq max neval
original 476.172 536.9770 567.86559 554.8600 580.0530 8054.805 1000
strsplit 366.071 422.3660 448.69815 445.3810 469.6410 798.389 1000
gregexpr 302.622 345.2325 423.08347 360.3170 378.0455 9082.416 1000
stringi 112.589 135.2940 149.82411 144.6245 155.1990 3910.770 1000
raw 58.161 71.5340 83.57614 77.1330 82.1090 6249.642 1000
tabulate 18.039 29.8575 35.20816 36.3890 40.7430 72.779 1000
Why the weird calculation?
This odd formula helps identify some plainly-looking fake identifiers entered by the user. For example, some non-creative users (I'm a guilty one as well) fill out same digits for their phone numbers. Frequently, in data analysis, it would be better to have no phone number at all than a fake phone number that changes from one dataset to another. Naturally, if there is a check-digit, it would be an additional easy validation.
回答1:
Using charToRaw to count digits in string:
# To count only digits in string, filter out ASCii codes for numbers from 0 to 9 which is 48 to 57 according to https://ascii.cl/
# You need to add na.rm = TRUE in case some of your strings contain only one digit
txt = 'wow:011 test 234567, abc=8951111111111aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa'
max(tabulate(as.integer(charToRaw(txt)))[48:57], na.rm = TRUE)
#[1] 12
txt='22222222222'
max(tabulate(as.integer(charToRaw(txt)))[48:57], na.rm = TRUE)
#[1] 11
@Andrew already did benchmarking test which proves that using charToRaw is fastest approach to count digits in string.
If you do not care about the digit and just want to count most frequent character/digit then you just remove filtering ASCII codes [48:57].
txt = 'wow:011 test 234567, abc=8951111111111aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa'
max(tabulate(as.integer(charToRaw(txt))))
#[1] 32
txt='22222222222'
max(tabulate(as.integer(charToRaw(txt))))
#[1] 11
回答2:
max(table(unlist(strsplit(gsub("\\D+", "", txt), ""))))
#OR
max(sapply(0:9, function(d) sum(unlist(gregexpr(d, txt)) > 0)))
#[1] 12
Or if you do care about the digit
with(rle(sort(unlist(strsplit(gsub("\\D+", "", txt), "")))),
setNames(c(max(lengths)), values[which.max(lengths)]))
# 1
#12
library(microbenchmark)
set.seed(42)
t = paste(sample(c(letters, 0:9), 1e5, TRUE), collapse = "")
microbenchmark(original = max(sapply(0:9, function(i) nchar(gsub(paste0('[^',i,']'), '', t)))),
strsplit = max(table(unlist(strsplit(gsub("\\D+", "", t), "")))),
gregexpr = max(sapply(0:9, function(d) sum(unlist(gregexpr(d, t)) > 0))))
#Unit: milliseconds
# expr min lq mean median uq max neval cld
# original 215.371764 220.862807 233.368696 228.757529 239.809292 308.94393 100 c
# strsplit 11.224226 11.856327 12.956749 12.320586 12.893789 30.61072 100 b
# gregexpr 7.542871 7.958818 8.680391 8.302971 8.728735 13.79921 100 a
回答3:
Building on Santosh's approach, this is significantly faster than the other options...
max(tabulate(as.integer(charToRaw(txt)))[48:57]) #48:57 picks out ASCII digits
library(microbenchmark)
set.seed(42)
t = paste(sample(c(letters, 0:9), 1e5, TRUE), collapse = "")
microbenchmark(original = max(sapply(0:9, function(i) nchar(gsub(paste0('[^',i,']'), '', t)))),
strsplit = max(table(unlist(strsplit(gsub("\\D+", "", t), "")))),
gregexpr = max(sapply(0:9, function(d) sum(unlist(gregexpr(d, t)) > 0))),
tabulate = max(tabulate(as.integer(charToRaw(t)))[48:57]))
Unit: milliseconds
expr min lq mean median uq max neval
original 807.947235 860.112901 1169.744733 935.169003 1154.057709 3513.1401 100
strsplit 34.100444 36.453163 55.457896 42.881400 58.208820 390.1453 100
gregexpr 27.205510 29.333569 42.616817 33.146572 49.840566 246.9001 100
tabulate 1.189702 1.208321 2.150022 1.226319 1.297068 37.4300 100
来源：https://stackoverflow.com/questions/47516752/fast-count-of-digits-in-a-string-in-r