如何找出瓶頸？

撰寫高效 R 程式碼

Colin Gillespie

Jumping Rivers & Newcastle University

程式剖析（profiling）

基本概念是：

執行程式碼
每隔數毫秒記錄當前執行內容
Rprof() 內建於 R，正是做這件事
- 不太好用
改用 profvis

IMDB 資料集

來自 ggplot2movies 套件

data(movies, package = "ggplot2movies")
dim(movies)

58788    24

資料框：約 60,000 列、24 欄
每列對應一部電影

Braveheart

braveheart = movies[7288,]

Year	Length	Rating
1995	177	8.3

範例：Braveheart

# Load data
data(movies,
     package = "ggplot2movies")
braveheart <- movies[7288,]
movies <- movies[movies$Action==1,]

plot(movies$year, movies$rating,
     xlab = "Year", ylab = "Rating")

# local regression line
model <- loess(rating ~ year,
               data = movies)
j <- order(movies$year)
lines(movies$year[j],
      model$fitted[j],
      col = "forestgreen")

points(braveheart$year,
       braveheart$rating,
       pch = 21,
       bg = "steelblue")

Profvis

RStudio 整合了 profvis 的剖析支援
- 反白選取要剖析的程式碼
- Profile -> Profile Selected lines

命令列

library("profvis")

profvis({

data(movies, package = "ggplot2movies") # Load data
braveheart <- movies[7288,]
movies <- movies[movies$Action == 1,]
plot(movies$year, movies$rating, xlab = "Year", ylab="Rating")
model <- loess(rating ~ year, data = movies) # loess regression line
j <- order(movies$year)
lines(movies$year[j], model$fitted[j], col="forestgreen", lwd=2)
points(braveheart$year, braveheart$rating,
     pch = 21, bg = "steelblue", cex = 3)

})

你認為哪一行最慢？

一起來練習吧！

撰寫高效 R 程式碼