R根据其参考列将特定列从一个数据帧合并到另一数据帧

Question

我是初学者R用户。我有两个巨大的数据框，我想在hkdata.2处添加一个名为Vaccine的新列，该数据是根据hkdata.2的2个参考列（hhID和成员）从另一个DF遵从性获取的，有人可以帮我吗？

hkdata.2
hhID    member  T0  delta   X_hh    X_fm    ILI age
1          1    7      0    0       0        0  44
1          2    7      0    0       0        0  36
2          1    8      0    1       0        0  39
2          2    8      0    1       0        0  39

adherence
hhID member mask soap vaccine
1      0      1    0    1   
1      1      1    1    1
1      2      0    0    1
2      0      1    0    0
2      1      0    0    0
2      2      1    0    1

所以最后我可以得到这样的东西。在hkdata.2中增加了一个称为疫苗的列

hkdata.2
hhID    member  T0  delta   X_hh    X_fm    ILI age vaccine
1          1    7      0    0       0        0  44    1
1          2    7      0    0       0        0  36    1
2          1    8      0    1       0        0  39    0
2          2    8      0    1       0        0  39    1

Answer 1

更新：使用v1.9.6的on=语法。有关旧代码，请参见历史记录。

require(data.table) # v1.9.6+
setDT(hkdata.2)[setDT(adherence), vaccine := i.vaccine, on=c("hhID", "member")]
#    hhID member T0 delta X_hh X_fm ILI age vaccine
# 1:    1      1  7     0    0    0   0  44       1
# 2:    1      2  7     0    0    0   0  36       1
# 3:    2      1  8     0    1    0   0  39       0
# 4:    2      2  8     0    1    0   0  39       1

1. setDT通过引用将data.frame转换为data.table 。

2. 对由指定的列执行联接on=。联接中使用的是您需要的Note that this join is both a) fast *and* b) memory efficient. a) *fast* because they're binary search based joins, and no copy is being made here at all. The疫苗column is directly added to yourhkdata.2 疫苗data.table. b) *memory efficient* because only the column，而不是其他列（对于非常大的数据集，这特别好）。

这是一个基准，假设每个100,000 hhID秒和200 member秒hhID：

require(data.table) # v1.9.6
require(dplyr) # 0.4.3.9000

set.seed(98192L)
N = 40e6 # 40 million rows
hkdata.2 = data.frame(hhID   = rep(1:1e5, each=200), 
                      member = 1:200,
                      T0     = sample(10),
                      delta  = sample(0:1), 
                      X_hh   = sample(0:1), 
                      X_fm   = sample(0:1), 
                      ILI    = sample(0:1),
                      age    = sample(30:100, N/2, TRUE))

# let's go with 100,000 hhIDs and 400 members here:
adherence = data.frame(hhID    = rep(1:1e5, each=400), 
                       member  = 1:400,
                       mask    = sample(0:1),
                       soap    = sample(0:1),
                       vaccine = sample(0:1))

## dplyr timing
system.time(ans1 <- left_join(hkdata.2, select(adherence, -soap, -mask)))
#   user  system elapsed
# 16.977   2.163  19.605
## data.table timing
system.time(setDT(hkdata.2)[setDT(adherence), vaccine := i.vaccine, on=c("hhID", "member")])
#   user  system elapsed
#  1.186   0.233   1.427

的峰值内存使用量dplyr为4.7GB，花费了19.6秒完成，而data.table花费了1.4秒，峰值内存使用量为2.2GB。

简介：data.table在这里，速度提高了约14倍，内存效率提高了约2.1倍。