在向量中查找第一个 TRUE 值的更快方法 [英] Faster way to find the first TRUE value in a vector

问题描述

在一个函数中，我经常需要使用如下代码:

In one function I very often need to use code like:

which(x==1)[1]
which(x>1)[1]
x[x>10][1]

其中 x 是一个数值向量.summaryRprof() 表明我在关系运算符上花费了 >80% 的时间.我想知道是否有一个函数只在达到第一个 TRUE 值之前进行比较以加速我的代码.For 循环比上面提供的选项慢.

where x is a numeric vector. summaryRprof() shows that I spend >80% of the time on relational operators. I wonder if there is a function that does comparison only till the first TRUE value is reached to speed up my code. For-loop is slower than the options provided above.

推荐答案

Base R 提供了 Position 和 Find 分别用于定位第一个索引和值，对于它们谓词返回真值.这些高阶函数在第一次命中时立即返回.

Base R provides Position and Find for locating the first index and value, respectively, for which a predicate returns a true value. These higher-order functions return immediately upon the first hit.

f<-function(x) {
  r<-vector("list",3)
  r[[1]]<-which(x==1)[1]
  r[[2]]<-which(x>1)[1]
  r[[3]]<-x[x>10][1]
  return(r)
}

p<-function(f,b) function(a) f(a,b)
g<-function(x) {
  r<-vector("list",3)
  r[[1]]<-Position(p(`==`,1),x)
  r[[2]]<-Position(p(`>`,1),x)
  r[[3]]<-Find(p(`>`,10),x)
  return(r)
}

相对性能在很大程度上取决于相对于谓词成本与 Position/Find 开销的早期发现命中概率.

The relative performance depends greatly on the probability of finding a hit early relative to the cost of the predicate vs the overhead of Position/Find.

library(microbenchmark)
set.seed(1)
x<-sample(1:100,1e5,replace=TRUE)
microbenchmark(f(x),g(x))

Unit: microseconds
 expr      min        lq     mean    median        uq      max neval cld
 f(x) 5034.283 5410.1205 6313.861 5798.4780 6948.5675 26735.52   100   b
 g(x)  587.463  650.4795 1013.183  734.6375  950.9845 20285.33   100  a

y<-rep(0,1e5)
microbenchmark(f(y),g(y))

Unit: milliseconds
 expr        min         lq       mean     median         uq        max neval cld
 f(y)   3.470179   3.604831   3.791592   3.718752   3.866952   4.831073   100  a 
 g(y) 131.250981 133.687454 137.199230 134.846369 136.193307 177.082128   100   b

这篇关于在向量中查找第一个 TRUE 值的更快方法的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！