通过在数据 table 中动态创建列名进行计算

Question

我在获取数据时遇到问题 table 通过从作为变量的其他列名称计算它们的值来创建 28 个新列，同时满足某些条件。

我要实现的是区分四种情况：

现任者 (ic) 和新来者 (nc) == 0 => 期望结果 = NA：这是第一个 if 子句，它有效。
Fields ic are > 0, but no nc present (0) => desired result = value of ic_col (i.e. the number of newcomers): 这是第二个 if 子句，它不起作用。
字段 ic == 0，但 nc 存在 (>0) => 期望的结果 = 0：这是第三个 if 子句并且有效。
字段 ic > 1 和 nc > 1（= 两者都存在）=> 期望的结果 = 两个数字的比率，即 ic_col / nc_col。这是第四个 if 子句，不起作用。

如您所见，我尝试了很多不同的变体（如果第 2 条试图以一种 DT 方式进行），如果第 4 条试图事先获取值（当然，这不起作用，因为 my_ic_value 中包含所有数千个值，而不是只有一行）。我尝试失败的其他事情是

all <- all[, ratio_col:= get(ic_col) / get(nc_col)), by=.(m_island_id_1, m_owner_id_1)]

同样只使用列名是行不通的，因为它们是字符串，不能被分割:

all <- all[, ratio_col:= (ic_col / nc_col), by=.(m_island_id_1, m_owner_id_1)]

我希望问题有点清楚，我期待所有 data.table 专家向我展示我的代码是多么混乱，以及它可以多么短和快速地完成！ ;-)

感谢您的帮助！

这里是一些数据（很小的一部分）：

structure(list(m_island_id_1 = c("020d49b9580f071075cfb6f9da7a426669a56d2d", 
"020d49b9580f071075cfb6f9da7a426669a56d2d", "020d49b9580f071075cfb6f9da7a426669a56d2d"
), m_owner_id_1 = c("01d96c16a2caf720617b266ae5d053243cef0920", 
"4e25ef4cf04f023d96134856098b8104fa6a6add", "7818c5081257a837e7cfad2c1aba4d90dcd18a69"
), Sum_nc = c(2L, 2L, 2L), Sum_ic = c(4L, 4L, 4L), Sum_ic_2 = c(0L, 
0L, 0L), Sum_ic_3 = c(0L, 0L, 0L), Sum_ic_4 = c(0L, 0L, 0L), 
    Sum_ic_5 = c(0L, 0L, 0L), Sum_ic_6 = c(0L, 0L, 0L), Sum_ic_7 = c(0L, 
    0L, 0L), Sum_ic_8 = c(0L, 0L, 0L), Sum_ic_9 = c(0L, 0L, 0L
    ), Sum_ic_10 = c(0L, 0L, 0L), Sum_ic_11 = c(0L, 0L, 0L), 
    Sum_ic_12 = c(0L, 0L, 0L), Sum_ic_13 = c(0L, 0L, 0L), Sum_ic_14 = c(0L, 
    0L, 0L), Sum_ic_15 = c(0L, 0L, 0L), Sum_ic_16 = c(0L, 0L, 
    0L), Sum_ic_17 = c(1L, 1L, 1L), Sum_ic_18 = c(1L, 1L, 1L), 
    Sum_ic_19 = c(2L, 2L, 2L), Sum_ic_20 = c(0L, 0L, 0L), Sum_ic_21 = c(0L, 
    0L, 0L), Sum_ic_22 = c(0L, 0L, 0L), Sum_ic_23 = c(0L, 0L, 
    0L), Sum_ic_24 = c(0L, 0L, 0L), Sum_ic_25 = c(0L, 0L, 0L), 
    Sum_ic_26 = c(0L, 0L, 0L), Sum_ic_27 = c(0L, 0L, 0L), Sum_ic_28 = c(0L, 
    0L, 0L), Sum_nc_2 = c(0L, 0L, 0L), Sum_nc_3 = c(0L, 0L, 0L
    ), Sum_nc_4 = c(0L, 0L, 0L), Sum_nc_5 = c(0L, 0L, 0L), Sum_nc_6 = c(0L, 
    0L, 0L), Sum_nc_7 = c(0L, 0L, 0L), Sum_nc_8 = c(0L, 0L, 0L
    ), Sum_nc_9 = c(0L, 0L, 0L), Sum_nc_10 = c(0L, 0L, 0L), Sum_nc_11 = c(0L, 
    0L, 0L), Sum_nc_12 = c(0L, 0L, 0L), Sum_nc_13 = c(0L, 0L, 
    0L), Sum_nc_14 = c(0L, 0L, 0L), Sum_nc_15 = c(0L, 0L, 0L), 
    Sum_nc_16 = c(1L, 1L, 1L), Sum_nc_17 = c(0L, 0L, 0L), Sum_nc_18 = c(1L, 
    1L, 1L), Sum_nc_19 = c(0L, 0L, 0L), Sum_nc_20 = c(0L, 0L, 
    0L), Sum_nc_21 = c(0L, 0L, 0L), Sum_nc_22 = c(0L, 0L, 0L), 
    Sum_nc_23 = c(0L, 0L, 0L), Sum_nc_24 = c(0L, 0L, 0L), Sum_nc_25 = c(0L, 
    0L, 0L), Sum_nc_26 = c(0L, 0L, 0L), Sum_nc_27 = c(0L, 0L, 
    0L), Sum_nc_28 = c(0L, 0L, 0L), inhabited = c(TRUE, TRUE, 
    FALSE), Sum_time_ic = c(3L, 1L, 0L), groupsize = c(2L, 2L, 
    2L), Ratio_nc_to_ic_16 = c(0, 0, 0), Ratio_nc_to_ic_17 = c(TRUE, 
    TRUE, TRUE), Ratio_nc_to_ic_18 = c(1, 1, 1), Ratio_nc_to_ic_19 = c(TRUE, 
    TRUE, TRUE), ratio_col = c(NaN, NaN, NaN), Ratio_nc_to_ic = c(NA, 
    NA, NA), Ratio_nc_to_ic_ = c(2, 2, 2), Ratio_nc_to_ic_2 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_3 = c(NaN, NaN, NaN), Ratio_nc_to_ic_4 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_5 = c(NaN, NaN, NaN), Ratio_nc_to_ic_6 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_7 = c(NaN, NaN, NaN), Ratio_nc_to_ic_8 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_9 = c(NaN, NaN, NaN), Ratio_nc_to_ic_10 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_11 = c(NaN, NaN, NaN), Ratio_nc_to_ic_12 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_13 = c(NaN, NaN, NaN), Ratio_nc_to_ic_14 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_15 = c(NaN, NaN, NaN), Ratio_nc_to_ic_20 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_21 = c(NaN, NaN, NaN), Ratio_nc_to_ic_22 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_23 = c(NaN, NaN, NaN), Ratio_nc_to_ic_24 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_25 = c(NaN, NaN, NaN), Ratio_nc_to_ic_26 = c(NaN, 
    NaN, NaN), Ratio_nc_to_ic_27 = c(NaN, NaN, NaN), Ratio_nc_to_ic_28 = c(NaN, 
    NaN, NaN)), row.names = c(NA, -3L), class = c("data.table", 
"data.frame"), .internal.selfref = <pointer: 0x000001f487d51ef0>)

这是无效代码：

for (j in 2:28){
  ic_col <- paste0("Sum_ic_", j)
  nc_col <- paste0("Sum_nc_", j)
  ratio_col <- paste0("Ratio_nc_to_ic_", j)
  my_ic_value <- all[,get(ic_col)]
  my_nc_value <- all[,get(nc_col)]

  if ((ic_col == 0) & (nc_col == 0)) {
    all <- all[, ratio_col:=  NA, by=.(m_island_id_1, m_owner_id_1)] # no incumbents, no newcomers
  }
  if ((ic_col > 0) & (nc_col == 0)) {
    all <- all[, .(ratio_col=  ic_col), by=.(m_island_id_1, m_owner_id_1)] # Incumbents without newcomers
  }
  if ((ic_col == 0) & (nc_col > 0)) {
    all <- all[, ratio_col:=  0, by=.(m_island_id_1, m_owner_id_1)] # Newcomers without incumbents
  }
  if ((ic_col >= 1) & (nc_col >= 1)) {
    all <- all[, ratio_col:= (my_ic_value / my_nc_value), by=.(m_island_id_1, m_owner_id_1)] # Newcomers with incumbents
  }
}

Answer 1

好像要按行进行计算，所以你可以去掉by（如果我错了请指正）。而使用矩阵运算可以覆盖后两种情况。然后使用一些索引来处理情况 1 和 2，如下所示：

DT[, (ratio_cols) := {
    icmat <- as.matrix(.SD[, mget(ic_cols)])
    ncmat <- as.matrix(.SD[, mget(nc_cols)])

    m <- icmat / ncmat

    #when nc_col==0
    m[is.infinite(m)] <- icmat[is.infinite(m)]

    #when ic_col==0 & nc_col==0
    m[icmat==0 & ncmat==0] <- NA_real_

    as.data.table(m)
}]

数据：

#where dat is your sample data and ratio cols are removed here
DT <- dat[, .SD, .SDcols=m_island_id_1:groupsize]

#clean up names
rename <- c("Sum_nc","Sum_ic")
setnames(DT, rename, paste0(rename, "_1"))

#input ic, nc and output ratio columns
jcols <- 1:28
ic_cols <- paste0("Sum_ic_", jcols)
nc_cols <- paste0("Sum_nc_", jcols)
ratio_cols <- paste0("Ratio_nc_to_ic_", jcols)

通过在数据 table 中动态创建列名进行计算

Calculate with creating column names in data table on the fly

r

calculated-columns

data.table