R | data.table

• data.table
  https://github.com/Rdatatable/data.table
  https://www.rstudio.com/resources/cheatsheets/
  知乎 | 张敬信 | 【R语言新书】2.7 数据处理神器：data.table包
  知乎 | 老俊俊 | data.table 让你高效快速地处理数据

“data.table 高度抽象的语法无疑增加了学习成本，但它的高效性能和处理大数据能力，使得非常有必要学习它。当然，读者如果既想要 data.table 的高性能，又想要 tidyverse 的整洁语法，也可以借助一些衔接二者的中间包，如 dtplyr, tidyfst 等。”

• dtplyr
  https://github.com/tidyverse/dtplyr
  https://dtplyr.tidyverse.org/

• tidyfst
  https://github.com/hope-data-science/tidyfst
  知乎 | 黄天元 | R语言高效数据框操作：tidyfst （注意有专栏）
  知乎 | 黄小伟 | R最快且比dplyr最高效的大数据处理R包：tidyfst

---

创建

• data.table():
• as.data.table()

读取

• fread("file.csv")
select = c("a", "b")): 读取指定的列

写出

• fwrite(dt, "file.csv")

行操作

• dt[1:2,]
• dt[a > 5,]
• dt[, c := 1:.N, by = b]
• dt[, c := shift(a, 1), by = b]
• dt[, c := shift(a, 1, type = "lead"), by = b]

>、<
>= <=
is.na()
!is.na()
%in%
|、&、!
%like%
%between%

列操作

• dt[, c(2)]
• dt[, .(b, c)]
• dt[, .(x = sum(a))]
• dt[, c := 1+2]
• dt[,`:=`(c = 1, d = 2)]
• dt[, c := NULL]
• dt[, b := as.integer(b)]
• dt[, lapply(.SD, mean). SDcols = c("a", "b")]
• cols <- c("a")
dt[, paste0(cols , "_m") := lapply(.SD, mean)]

分组

• dt[, j, by = .(a)]
• dt[, j, keyby = .(a)]
• dt[, .(c = sum(b)), by = a)]
• dt[, c := sum(b), by = a]
• dt[, .SD[1], by = a]
• dt[, .SD[.N], by = a]

• dt[...][...]

函数

• setorder(dt, a, -b)
• unique[dt, by = c("a", "b")]: 去重
• uniqueN(dt, by = c("a", "b")): 计数
• setnames(dt, c("a", "b"), c("x", "y")): 重命名
• setkey(dt, a, b)

data.table中 以set为前缀的函数和操作符:=不需要<-就可以改变数据。
例如，setDT(df) 等同于 df < - as.data.table(df)。

合并

• dt_a[dt_b, on = .(b = y)]
• dt_a[dt_b, on = .(b = y, c > z)]
• rbind(dt_a, dt_b)
• cbind(dt_a, dt_b)

重塑

• dcast(): 长变宽
• melt(): 宽变长

dcast(dt,
      id - y,
      value.var = c("a", "b"))

melt(dt,
     id.vars = c("id"),
     measure.vars = patterns("^a", "^b"),
     variable.name = "y",
     value.name = c("a", "b"))