字符串向量中的高效模式检测

时间:2019-02-15 22:08:40

标签: r performance pattern-matching stringr

我正在寻找对this question提出的问题最有效的解决方案:假设您有一个向量v的字符串:

set.seed(314159)
library(stringi)
library(stringr)

v <- stringi::stri_rand_strings(10000, 4, pattern = "[A-Z]")

head(v)
#> [1] "FQGK" "YNQH" "IMNJ" "WUFU" "BBAR" "BZUH"

我想有效地返回一个逻辑,该逻辑表示给定的模式(例如"FOO")是否与v中的任何字符串匹配。预期功能将像这样工作:

detect("FOO")
#> FALSE
detect("BAR")
#> TRUE

有几种方法可以使用基本的grep函数或使用stringr::str_detect来进行此操作,但是每种方法都涉及首先在v的每个元素上匹配一个正则表达式,最多可以执行9,999个不必要的测试在我的例子中。找到单个匹配项后,有效的解决方案将停止评估。

对于每个解决方案detect.#,我通过将其应用于所有三个字母组合c进行基准测试:


c <- combn(LETTERS,3, FUN = function(x){paste(x, collapse = '')})
head(c)
#> [1] "ABC" "ABD" "ABE" "ABF" "ABG" "ABH"

可能的解决方案

我想出了几种可能的解决方案。首先,循环v,以便在找到匹配项后不再进行不必要的模式匹配。如您所见,这是一个糟糕的主意,有很多开销:

detect.1 <- function(pattern){
  for (i in 1:length(v)){
    if (length(grep(pattern, v[i]))){return(TRUE)}
  }
  return(FALSE)
}

接下来,我们可以使用any()grepl()stringr::str_detect()的组合,但是随后我们进行了不必要的匹配测试:

#str_detect() from stringr
detect.2 <- function(pattern){
  any(str_detect(v, pattern) )
}

# any() and grepl()
detect.3 <- function(pattern){
  any(grepl(pattern, v))
}

最后,如果我们知道一个字符从未出现在pattern中,我们可以将v折叠成一个由该字符分隔的单个字符串。然后一个grep就足够了:

#collapse to long string
v_pasted <- paste(v, collapse = '_')
detect.4 <- function(pattern){
  isTRUE(as.logical(grep(pattern, v_pasted)))
}

基准

(已更新为使用bench::mark()

det1 <- expression(data.frame(c, "inV" = I(lapply(c, FUN = detect.1))))
det2 <- expression(data.frame(c, "inV" = I(lapply(c, FUN = detect.2))))
det3 <- expression(data.frame(c, "inV" = I(lapply(c, FUN = detect.3))))
det4 <- expression({
  v_pasted <- paste(v, collapse = '_')
  data.frame(c, "inV" = I(lapply(c, FUN = detect.4)))
})

bench::mark(
  eval(det1),
  eval(det2),
  eval(det3),
  eval(det4),
  iterations = 5,
  relative = TRUE
)
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 4 x 10
#>   expression   min  mean median   max `itr/sec` mem_alloc  n_gc n_itr
#>   <chr>      <dbl> <dbl>  <dbl> <dbl>     <dbl>     <dbl> <dbl> <dbl>
#> 1 eval(det1) 76.9  77.0   76.8  77.2        1        1      Inf     1
#> 2 eval(det2)  4.02  4.03   4.04  4.05      19.1    735.     Inf     1
#> 3 eval(det3)  2.77  2.79   2.79  2.80      27.6    735.     Inf     1
#> 4 eval(det4)  1     1      1     1         77.0      1.22   NaN     1

greplstr_detect快得多。粘贴方法是最快的,但是要求您具有一个分隔符,该分隔符不会出现在可能的搜索模式中。有没有我想念的更快的选择?

1 个答案:

答案 0 :(得分:1)

stringi软件包中的此功能应该更快:

any(stri_detect_fixed(v, pattern, max_count = 1))

长凳:

require(stringi)
detect.m <- function(pattern){
  any(stri_detect_fixed(v, pattern, max_count = 1))
}

detm <- expression(data.frame(c, "inV" = I(lapply(c, FUN = detect.m))))
r <- bench::mark(
  # eval(det1),
  eval(det2),
  eval(det3),
  eval(det4),
  eval(detm),
  iterations = 5,
  relative = TRUE
)
r[, 1:10]
#   expression   min  mean median   max `itr/sec` mem_alloc  n_gc n_itr total_time
#    <chr>      <dbl> <dbl>  <dbl> <dbl>     <dbl>     <dbl> <dbl> <dbl>      <dbl>
# 1 eval(det2)  4.83  5.39   5.02  5.94      1         600.     9     1       5.39
# 2 eval(det3)  3.85  3.69   3.80  3.31      1.46      600.    10     1       3.69
# 3 eval(det4)  1.35  1.32   1.36  1.20      4.08        1      1     1       1.32
# 4 eval(detm)  1     1      1     1         5.39      600.     9     1       1   

更大的基准

# lets create larger test case for better comparison:

a <- expand.grid(lapply(1:5, function(x) LETTERS))
a <- do.call(paste0, a)
f10 <- a[10] # lets search for 10th element
last <- a[length(a)] # and last
length(a)
length(unique(a))

v <- a

detm <- function(pattern){
  any(stri_detect_fixed(v, pattern, max_count = 1))
}

det4 <- function(pattern){
  # should include paste
  v_pasted <- paste(v, collapse = '_')
  # isTRUE(as.logical(grep(pattern, v_pasted)))
  isTRUE(grepl(pattern, v_pasted, fixed = T)) # faster
}

system.time(detm(last)) # 0.74
system.time(detm(f10)) #  0.33

system.time(det4(last)) # 3.38
system.time(det4(f10)) #  3.08