如果我有一些R列表mylist，你可以像这样添加一个obj项:
mylist[[length(mylist)+1]] <- obj
但肯定有一些更紧凑的方式。当我刚在R工作时，我试着像这样写lappend():
lappend <- function(lst, obj) {
lst[[length(lst)+1]] <- obj
return(lst)
}
但是，由于R的按名调用语义(lst在调用时被有效地复制，因此对lst的更改在lappend()的作用域之外是不可见的)，这当然是行不通的。我知道您可以在R函数中进行环境入侵，从而超出函数的作用域并改变调用环境，但对于编写一个简单的附加函数来说，这似乎是一个巨大的打击。
有谁能提出一个更漂亮的方法吗?如果它对向量和列表都适用，那就更好了。
当前回答
还有一个列表。从rlist中追加(链接到文档)
require(rlist)
LL <- list(a="Tom", b="Dick")
list.append(LL,d="Pam",f=c("Joe","Ann"))
这非常简单和有效。
其他回答
在其他答案中，只有列表方法会导致O(1)个追加，但它会导致深度嵌套的列表结构，而不是简单的单个列表。我使用了下面的数据结构，它们支持O(1)(平摊)追加，并允许结果转换回一个普通列表。
expandingList <- function(capacity = 10) {
buffer <- vector('list', capacity)
length <- 0
methods <- list()
methods$double.size <- function() {
buffer <<- c(buffer, vector('list', capacity))
capacity <<- capacity * 2
}
methods$add <- function(val) {
if(length == capacity) {
methods$double.size()
}
length <<- length + 1
buffer[[length]] <<- val
}
methods$as.list <- function() {
b <- buffer[0:length]
return(b)
}
methods
}
and
linkedList <- function() {
head <- list(0)
length <- 0
methods <- list()
methods$add <- function(val) {
length <<- length + 1
head <<- list(head, val)
}
methods$as.list <- function() {
b <- vector('list', length)
h <- head
for(i in length:1) {
b[[i]] <- head[[2]]
head <- head[[1]]
}
return(b)
}
methods
}
使用方法如下:
> l <- expandingList()
> l$add("hello")
> l$add("world")
> l$add(101)
> l$as.list()
[[1]]
[1] "hello"
[[2]]
[1] "world"
[[3]]
[1] 101
这些解决方案可以扩展为支持所有列表相关操作的完整对象，但这仍将作为读者的练习。
命名列表的另一种变体:
namedExpandingList <- function(capacity = 10) {
buffer <- vector('list', capacity)
names <- character(capacity)
length <- 0
methods <- list()
methods$double.size <- function() {
buffer <<- c(buffer, vector('list', capacity))
names <<- c(names, character(capacity))
capacity <<- capacity * 2
}
methods$add <- function(name, val) {
if(length == capacity) {
methods$double.size()
}
length <<- length + 1
buffer[[length]] <<- val
names[length] <<- name
}
methods$as.list <- function() {
b <- buffer[0:length]
names(b) <- names[0:length]
return(b)
}
methods
}
基准
使用@phonetagger的代码(基于@Cron Arconis的代码)进行性能比较。我还添加了一个better_env_as_container，并对env_as_container_进行了一些更改。原来的env_as_container_被破坏了，实际上并没有存储所有的数字。
library(microbenchmark)
lPtrAppend <- function(lstptr, lab, obj) {lstptr[[deparse(lab)]] <- obj}
### Store list inside new environment
envAppendList <- function(lstptr, obj) {lstptr$list[[length(lstptr$list)+1]] <- obj}
env2list <- function(env, len) {
l <- vector('list', len)
for (i in 1:len) {
l[[i]] <- env[[as.character(i)]]
}
l
}
envl2list <- function(env, len) {
l <- vector('list', len)
for (i in 1:len) {
l[[i]] <- env[[paste(as.character(i), 'L', sep='')]]
}
l
}
runBenchmark <- function(n) {
microbenchmark(times = 5,
env_with_list_ = {
listptr <- new.env(parent=globalenv())
listptr$list <- NULL
for(i in 1:n) {envAppendList(listptr, i)}
listptr$list
},
c_ = {
a <- list(0)
for(i in 1:n) {a = c(a, list(i))}
},
list_ = {
a <- list(0)
for(i in 1:n) {a <- list(a, list(i))}
},
by_index = {
a <- list(0)
for(i in 1:n) {a[length(a) + 1] <- i}
a
},
append_ = {
a <- list(0)
for(i in 1:n) {a <- append(a, i)}
a
},
env_as_container_ = {
listptr <- new.env(hash=TRUE, parent=globalenv())
for(i in 1:n) {lPtrAppend(listptr, i, i)}
envl2list(listptr, n)
},
better_env_as_container = {
env <- new.env(hash=TRUE, parent=globalenv())
for(i in 1:n) env[[as.character(i)]] <- i
env2list(env, n)
},
linkedList = {
a <- linkedList()
for(i in 1:n) { a$add(i) }
a$as.list()
},
inlineLinkedList = {
a <- list()
for(i in 1:n) { a <- list(a, i) }
b <- vector('list', n)
head <- a
for(i in n:1) {
b[[i]] <- head[[2]]
head <- head[[1]]
}
},
expandingList = {
a <- expandingList()
for(i in 1:n) { a$add(i) }
a$as.list()
},
inlineExpandingList = {
l <- vector('list', 10)
cap <- 10
len <- 0
for(i in 1:n) {
if(len == cap) {
l <- c(l, vector('list', cap))
cap <- cap*2
}
len <- len + 1
l[[len]] <- i
}
l[1:len]
}
)
}
# We need to repeatedly add an element to a list. With normal list concatenation
# or element setting this would lead to a large number of memory copies and a
# quadratic runtime. To prevent that, this function implements a bare bones
# expanding array, in which list appends are (amortized) constant time.
expandingList <- function(capacity = 10) {
buffer <- vector('list', capacity)
length <- 0
methods <- list()
methods$double.size <- function() {
buffer <<- c(buffer, vector('list', capacity))
capacity <<- capacity * 2
}
methods$add <- function(val) {
if(length == capacity) {
methods$double.size()
}
length <<- length + 1
buffer[[length]] <<- val
}
methods$as.list <- function() {
b <- buffer[0:length]
return(b)
}
methods
}
linkedList <- function() {
head <- list(0)
length <- 0
methods <- list()
methods$add <- function(val) {
length <<- length + 1
head <<- list(head, val)
}
methods$as.list <- function() {
b <- vector('list', length)
h <- head
for(i in length:1) {
b[[i]] <- head[[2]]
head <- head[[1]]
}
return(b)
}
methods
}
# We need to repeatedly add an element to a list. With normal list concatenation
# or element setting this would lead to a large number of memory copies and a
# quadratic runtime. To prevent that, this function implements a bare bones
# expanding array, in which list appends are (amortized) constant time.
namedExpandingList <- function(capacity = 10) {
buffer <- vector('list', capacity)
names <- character(capacity)
length <- 0
methods <- list()
methods$double.size <- function() {
buffer <<- c(buffer, vector('list', capacity))
names <<- c(names, character(capacity))
capacity <<- capacity * 2
}
methods$add <- function(name, val) {
if(length == capacity) {
methods$double.size()
}
length <<- length + 1
buffer[[length]] <<- val
names[length] <<- name
}
methods$as.list <- function() {
b <- buffer[0:length]
names(b) <- names[0:length]
return(b)
}
methods
}
结果:
> runBenchmark(1000)
Unit: microseconds
expr min lq mean median uq max neval
env_with_list_ 3128.291 3161.675 4466.726 3361.837 3362.885 9318.943 5
c_ 3308.130 3465.830 6687.985 8578.913 8627.802 9459.252 5
list_ 329.508 343.615 389.724 370.504 449.494 455.499 5
by_index 3076.679 3256.588 5480.571 3395.919 8209.738 9463.931 5
append_ 4292.321 4562.184 7911.882 10156.957 10202.773 10345.177 5
env_as_container_ 24471.511 24795.849 25541.103 25486.362 26440.591 26511.200 5
better_env_as_container 7671.338 7986.597 8118.163 8153.726 8335.659 8443.493 5
linkedList 1700.754 1755.439 1829.442 1804.746 1898.752 1987.518 5
inlineLinkedList 1109.764 1115.352 1163.751 1115.631 1206.843 1271.166 5
expandingList 1422.440 1439.970 1486.288 1519.728 1524.268 1525.036 5
inlineExpandingList 942.916 973.366 1002.461 1012.197 1017.784 1066.044 5
> runBenchmark(10000)
Unit: milliseconds
expr min lq mean median uq max neval
env_with_list_ 357.760419 360.277117 433.810432 411.144799 479.090688 560.779139 5
c_ 685.477809 734.055635 761.689936 745.957553 778.330873 864.627811 5
list_ 3.257356 3.454166 3.505653 3.524216 3.551454 3.741071 5
by_index 445.977967 454.321797 515.453906 483.313516 560.374763 633.281485 5
append_ 610.777866 629.547539 681.145751 640.936898 760.570326 763.896124 5
env_as_container_ 281.025606 290.028380 303.885130 308.594676 314.972570 324.804419 5
better_env_as_container 83.944855 86.927458 90.098644 91.335853 92.459026 95.826030 5
linkedList 19.612576 24.032285 24.229808 25.461429 25.819151 26.223597 5
inlineLinkedList 11.126970 11.768524 12.216284 12.063529 12.392199 13.730200 5
expandingList 14.735483 15.854536 15.764204 16.073485 16.075789 16.081726 5
inlineExpandingList 10.618393 11.179351 13.275107 12.391780 14.747914 17.438096 5
> runBenchmark(20000)
Unit: milliseconds
expr min lq mean median uq max neval
env_with_list_ 1723.899913 1915.003237 1921.23955 1938.734718 1951.649113 2076.910767 5
c_ 2759.769353 2768.992334 2810.40023 2820.129738 2832.350269 2870.759474 5
list_ 6.112919 6.399964 6.63974 6.453252 6.910916 7.321647 5
by_index 2163.585192 2194.892470 2292.61011 2209.889015 2436.620081 2458.063801 5
append_ 2832.504964 2872.559609 2983.17666 2992.634568 3004.625953 3213.558197 5
env_as_container_ 573.386166 588.448990 602.48829 597.645221 610.048314 642.912752 5
better_env_as_container 154.180531 175.254307 180.26689 177.027204 188.642219 206.230191 5
linkedList 38.401105 47.514506 46.61419 47.525192 48.677209 50.952958 5
inlineLinkedList 25.172429 26.326681 32.33312 34.403442 34.469930 41.293126 5
expandingList 30.776072 30.970438 34.45491 31.752790 38.062728 40.712542 5
inlineExpandingList 21.309278 22.709159 24.64656 24.290694 25.764816 29.158849 5
我已经添加了linkedList和expandingList和两者的内联版本。inlinedLinkedList基本上是list_的副本，但它也将嵌套结构转换回普通列表。除此之外，内联版本和非内联版本之间的差异是由于函数调用的开销。
expandingList和linkedList的所有变体都显示O(1)附加性能，基准时间随附加项的数量线性扩展。linkedList比expandingList慢，而且函数调用开销也是可见的。所以如果你真的需要所有你能得到的速度(并且想坚持R代码)，使用一个内联版本的expandingList。
我还研究了R的C实现，这两种方法都应该是O(1)追加任何大小，直到耗尽内存。
我也改变了env_as_container_，原始版本将存储索引“I”下的每一项，覆盖先前追加的项。我添加的better_env_as_container非常类似于env_as_container_，但没有离开的东西。两者都表现出O(1)性能，但它们的开销比链接/展开列表要大得多。
内存开销
In the C R implementation there is an overhead of 4 words and 2 ints per allocated object. The linkedList approach allocates one list of length two per append, for a total of (4*8+4+4+2*8=) 56 bytes per appended item on 64-bit computers (excluding memory allocation overhead, so probably closer to 64 bytes). The expandingList approach uses one word per appended item, plus a copy when doubling the vector length, so a total memory usage of up to 16 bytes per item. Since the memory is all in one or two objects the per-object overhead is insignificant. I haven't looked deeply into the env memory usage, but I think it will be closer to linkedList.
我对这里提到的方法做了一个小的比较。
n = 1e+4
library(microbenchmark)
### Using environment as a container
lPtrAppend <- function(lstptr, lab, obj) {lstptr[[deparse(substitute(lab))]] <- obj}
### Store list inside new environment
envAppendList <- function(lstptr, obj) {lstptr$list[[length(lstptr$list)+1]] <- obj}
microbenchmark(times = 5,
env_with_list_ = {
listptr <- new.env(parent=globalenv())
listptr$list <- NULL
for(i in 1:n) {envAppendList(listptr, i)}
listptr$list
},
c_ = {
a <- list(0)
for(i in 1:n) {a = c(a, list(i))}
},
list_ = {
a <- list(0)
for(i in 1:n) {a <- list(a, list(i))}
},
by_index = {
a <- list(0)
for(i in 1:n) {a[length(a) + 1] <- i}
a
},
append_ = {
a <- list(0)
for(i in 1:n) {a <- append(a, i)}
a
},
env_as_container_ = {
listptr <- new.env(parent=globalenv())
for(i in 1:n) {lPtrAppend(listptr, i, i)}
listptr
}
)
结果:
Unit: milliseconds
expr min lq mean median uq max neval cld
env_with_list_ 188.9023 198.7560 224.57632 223.2520 229.3854 282.5859 5 a
c_ 1275.3424 1869.1064 2022.20984 2191.7745 2283.1199 2491.7060 5 b
list_ 17.4916 18.1142 22.56752 19.8546 20.8191 36.5581 5 a
by_index 445.2970 479.9670 540.20398 576.9037 591.2366 607.6156 5 a
append_ 1140.8975 1316.3031 1794.10472 1620.1212 1855.3602 3037.8416 5 b
env_as_container_ 355.9655 360.1738 399.69186 376.8588 391.7945 513.6667 5 a
> LL<-list(1:4)
> LL
[[1]]
[1] 1 2 3 4
> LL<-list(c(unlist(LL),5:9))
> LL
[[1]]
[1] 1 2 3 4 5 6 7 8 9
事实上，c()函数有一个微妙之处。如果你有:
x <- list()
x <- c(x,2)
x = c(x,"foo")
如你所料，你将获得:
[[1]]
[1]
[[2]]
[1] "foo"
但是如果你添加一个x <- c(x, matrix(5,2,2))的矩阵，你的列表将有另外4个值为5的元素! 你最好做:
x <- c(x, list(matrix(5,2,2))
它适用于任何其他对象，你将获得预期的:
[[1]]
[1]
[[2]]
[1] "foo"
[[3]]
[,1] [,2]
[1,] 5 5
[2,] 5 5
最后，你的函数变成:
push <- function(l, ...) c(l, list(...))
它适用于任何类型的对象。你可以更聪明地去做:
push_back <- function(l, ...) c(l, list(...))
push_front <- function(l, ...) c(list(...), l)
这是一个非常有趣的问题，我希望我下面的想法可以为解决这个问题提供一种方式。这个方法给出了一个没有索引的平面列表，但是它有列表和反列表来避免嵌套结构。我不确定速度，因为我不知道如何基准。
a_list<-list()
for(i in 1:3){
a_list<-list(unlist(list(unlist(a_list,recursive = FALSE),list(rnorm(2))),recursive = FALSE))
}
a_list
[[1]]
[[1]][[1]]
[1] -0.8098202 1.1035517
[[1]][[2]]
[1] 0.6804520 0.4664394
[[1]][[3]]
[1] 0.15592354 0.07424637
