基于两个或多个变量的所有可能组合的子集 data.table

Subset data.table based on all possible combinations of two or more variables

我想根据某些变量是否全为正数、全为负数或介于两者之间的某种组合来对 data.frame 进行子集化。对于 n 个变量,这应该会导致 2^n 个可能的组合。

我认为 combn 可用于实现此目的,但我正在努力正确地做到这一点。

示例数据:

library(data.table)
dt <- data.table(x = runif(100, -1, 1), y = runif(100, -1, 1), z = runif(100, -1, 1))

我想要的:

dt[x < 0 & y < 0 z < 0, ]
dt[x < 0 & y < 0 z > 0, ]
dt[x < 0 & y > 0 z < 0, ]
dt[x < 0 & y > 0 z > 0, ]
dt[x > 0 & y < 0 z < 0, ]
dt[x > 0 & y < 0 z > 0, ]
dt[x > 0 & y > 0 z < 0, ]
dt[x > 0 & y > 0 z > 0, ]

到目前为止我尝试过的:

combinator <- function(z){
  cnames <- colnames(z)
  combinations <- t(combn(c(rep("<", ncol(z)), rep(">", ncol(z))),ncol(z)))

  retval <- t(sapply(1:nrow(combinations), function(p){
    sapply(1:ncol(z), function(q) paste(cnames[q], combinations[p,q], 0))
  }))

  return(apply(retval, 1, paste, collapse = " & "))
}

输出:

> l <- combinator(dt)
> l
 [1] "x < 0 & y < 0 & z < 0" "x < 0 & y < 0 & z > 0" "x < 0 & y < 0 & z > 0" "x < 0 & y < 0 & z > 0"
 [5] "x < 0 & y < 0 & z > 0" "x < 0 & y < 0 & z > 0" "x < 0 & y < 0 & z > 0" "x < 0 & y > 0 & z > 0"
 [9] "x < 0 & y > 0 & z > 0" "x < 0 & y > 0 & z > 0" "x < 0 & y < 0 & z > 0" "x < 0 & y < 0 & z > 0"
[13] "x < 0 & y < 0 & z > 0" "x < 0 & y > 0 & z > 0" "x < 0 & y > 0 & z > 0" "x < 0 & y > 0 & z > 0"
[17] "x < 0 & y > 0 & z > 0" "x < 0 & y > 0 & z > 0" "x < 0 & y > 0 & z > 0" "x > 0 & y > 0 & z > 0"

> l[1]
[1] "x < 0 & y < 0 & z < 0"

> subset(dt, eval(l[1]))
Error in subset.data.table(dt, eval(l[1])) : 
  'subset' must evaluate to logical

此外,如果以下显示我没有列出所有需要的组合:

> unique(l)
[1] "x < 0 & y < 0 & z < 0" "x < 0 & y < 0 & z > 0" 
[3] "x < 0 & y > 0 & z > 0" "x > 0 & y > 0 & z > 0"

输出应该有 8 个唯一结果,而不是上面显示的 4 个。

只需执行 dt[, sign_combi := do.call(paste, lapply(dt, sign))],您可以根据需要 splitby = 该列,例如 split(dt, dt$sign_combi)。尝试将代码粘贴在一起是个坏主意。

例如:

set.seed(47) # setting seed for reproducibility
dt <- data.table(x = runif(100, -1, 1), y = runif(100, -1, 1), z = runif(100, -1, 1))

# create combination column (you could keep it separate if you prefer)
dt[, sign_combi := do.call(paste, lapply(dt, sign))]

# split original data by sign combinations
result = split(dt, dt$sign_combi)

# list of 8 resulting data tables
length(result)
# [1] 8

# peaking at the first three rows of the first three tables:
lapply(head(result, 3), head, 3)
# $`-1 -1 -1`
#             x          y          z sign_combi
# 1: -0.5713038 -0.7103555 -0.6873705   -1 -1 -1
# 2: -0.1407803 -0.8371153 -0.3686299   -1 -1 -1
# 3: -0.6478446 -0.7629461 -0.7458949   -1 -1 -1
# 
# $`-1 -1 1`
#             x          y         z sign_combi
# 1: -0.8070969 -0.3952283 0.9212030    -1 -1 1
# 2: -0.1190934 -0.4969318 0.8082232    -1 -1 1
# 3: -0.6536104 -0.3280965 0.6880454    -1 -1 1
# 
# $`-1 1 -1`
#              x         y          z sign_combi
# 1: -0.78789241 0.8577848 -0.7586369    -1 1 -1
# 2: -0.04442825 0.4736388 -0.3354734    -1 1 -1
# 3: -0.22105744 0.3012645 -0.4160631    -1 1 -1