理解矢量化 [英] Understanding vectorisation

查看:220
本文介绍了理解矢量化的处理方法,对大家解决问题具有一定的参考价值,需要的朋友们下面随着小编来一起学习吧!

问题描述

我正在寻找一种将 R 中的大数字格式化为 2.3K 5.6M 。我在SO上找到了这个解决方案。原来,它显示了一些奇怪的行为,一些输入向量。

这是我想了解的 - $ / b>

< (302.456500093388,32553.3619756151,3323.71232001074,12065.4076372462,
0,6270.87962956305,383.337515655172,402.20778095643,19466.0204345063,$ b)pre $ #具有奇怪行为的测试向量
x< $ b 1779.05474064539,1467.09928489114,3786.27112222457,2080.08078309959,
51114.7097545816,51188.7710104291,59713.9414049798)

#格式化大数
comprss函数< - function(tx){
div < - findInterval(as.numeric(gsub(\\,,tx)),
c(1,1e3,1e6,1e9,1e12))
paste round(as.numeric(gsub(\\,,tx))/ 10 ^(3 *(div-1)),1),
c('','K' 'b','T')[div],sep ='')
}

#比较以下三个命令的输出
x
comprss(x)
sapply(x,comprss)

我们可以看到 comprss(x)产生了 0k 作为5 th 元素是奇怪的,但是 comprss(x [5])给了我们预期的结果。据我所知,在的主体中使用的所有函数, comprss 被矢量化。那么为什么我仍然需要 sapply 我的出路呢?

解决方案

这是一个从 pryr ::: print.bytes 改编的矢量化版本:





< (x,digits = 3){
grouping <-pmax(floor(log(abs(x), 1000)),0)
paste0(signif(x /(1000 ^ grouping),digits = digits),
c('','K','M','B','T' )[grouping + 1])$ ​​b $ b}

format_for_humans(10 ^ seq(0,12,2))
#> [1]110010K1M100M10B1T

x < - c(302.456500093388,32553.3619756151,3323.71232001074,12065.4076372462,
0 ,6270.87962956305,383.337515655172,402.20778095643,19466.0204345063,
1779.05474064539,1467.09928489114,3786.27112222457,2080.08078309959,
51114.7097545816,51188.7710104291,59713.9414049798)

format_for_humans(x)
#> [1]30232.6K3.32K12.1K06.27K383402
#> [9]19.5K1.78K1.47K3.79K2.08K51.1K51.2K59.7K

format_for_humans(x,digits = 1 )
#> [1]30030K3K10K06K40040020K2K1K
#> [12]4K2K50K50K60K


I was looking for a way to format large numbers in R as 2.3K or 5.6M. I found this solution on SO. Turns out, it shows some strange behaviour for some input vectors.

Here is what I am trying to understand -

# Test vector with weird behaviour
x <- c(302.456500093388, 32553.3619756151, 3323.71232001074, 12065.4076372462, 
  0, 6270.87962956305, 383.337515655172, 402.20778095643, 19466.0204345063, 
  1779.05474064539, 1467.09928489114, 3786.27112222457, 2080.08078309959, 
  51114.7097545816, 51188.7710104291, 59713.9414049798)

# Formatting function for large numbers
comprss <- function(tx) { 
  div <- findInterval(as.numeric(gsub("\\,", "", tx)), 
                      c(1, 1e3, 1e6, 1e9, 1e12) )
  paste(round( as.numeric(gsub("\\,","",tx))/10^(3*(div-1)), 1), 
        c('','K','M','B','T')[div], sep = '')
}

# Compare outputs for the following three commands
x
comprss(x)
sapply(x, comprss)

We can see that comprss(x) produces 0k as the 5th element which is weird, but comprss(x[5]) gives us the expected results. The 6th element is even weirder.

As far as I know, all the functions used in the body of comprss are vectorised. Then why do I still need to sapply my way out of this?

解决方案

Here's a vectorized version adapted from pryr:::print.bytes:

format_for_humans <- function(x, digits = 3){
    grouping <- pmax(floor(log(abs(x), 1000)), 0)
    paste0(signif(x / (1000 ^ grouping), digits = digits), 
           c('', 'K', 'M', 'B', 'T')[grouping + 1])
}

format_for_humans(10 ^ seq(0, 12, 2))
#> [1] "1"    "100"  "10K"  "1M"   "100M" "10B"  "1T"

x <- c(302.456500093388, 32553.3619756151, 3323.71232001074, 12065.4076372462, 
       0, 6270.87962956305, 383.337515655172, 402.20778095643, 19466.0204345063, 
       1779.05474064539, 1467.09928489114, 3786.27112222457, 2080.08078309959, 
       51114.7097545816, 51188.7710104291, 59713.9414049798)

format_for_humans(x)
#>  [1] "302"   "32.6K" "3.32K" "12.1K" "0"     "6.27K" "383"   "402"  
#>  [9] "19.5K" "1.78K" "1.47K" "3.79K" "2.08K" "51.1K" "51.2K" "59.7K"

format_for_humans(x, digits = 1)
#>  [1] "300" "30K" "3K"  "10K" "0"   "6K"  "400" "400" "20K" "2K"  "1K" 
#> [12] "4K"  "2K"  "50K" "50K" "60K"

这篇关于理解矢量化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!

查看全文
登录 关闭
扫码关注1秒登录
发送“验证码”获取 | 15天全站免登陆