向量子集化性能:名称与索引

时间:2013-12-02 09:58:55

标签: r

如果我有一个名为

的向量v
John       Murray     Lisa       Mike       Joe       Ann 
0.0832090  0.0475580 -0.2797860  0.1086225  0.0104590 -0.0028250 

v['Joe']v[4]的时间复杂度是多少?我想前者会采用O(log n),因为它应该涉及二进制搜索,但我仍然不确定后者是否为O(1)。

此外,结果是否概括为v是列表/数据框而不是原子向量的情况?

1 个答案:

答案 0 :(得分:7)

在名称查找的情况下,似乎是近似 O(n),即矢量扫描。您使用索引查找O(1)的猜想似乎是合理的......

#  Unique names for longish vector
nms <- apply( expand.grid( letters , letters , letters , letters ) , 1 , paste , collapse = "" )
length(nms)
#[1] 456976
length(unique(nms))
#[1] 456976

#  Start of names
head(nms)
#[1] "aaaa" "baaa" "caaa" "daaa" "eaaa" "faaa"

#  End of names
tail(nms)
#[1] "uzzz" "vzzz" "wzzz" "xzzz" "yzzz" "zzzz"

#  Large named vector
x <- setNames( runif( 456976 ) , nms )

#  Small named vector
y <- setNames( runif(26) , letters )

#  Timing information
require( microbenchmark )
bm <- microbenchmark( x['daaa'] , x[4] , x['vzzz'] , x[456972] , y['d'] , y[4] )
print( bm , order = 'median' , unit = 'relative' , digits = 3 )
#Unit: relative
#      expr min       lq   median       uq      max neval
# x[456972] NaN 1.00e+00     1.00     1.00    1.000   100
#      x[4] Inf 1.00e+00     1.33     1.07    0.957   100
#      y[4] NaN 5.01e-01     1.33     1.14    0.191   100
#    y["d"] Inf 1.00e+00     2.00     1.25    0.265   100
# x["vzzz"] Inf 6.57e+04 44412.24  9969.64 3439.154   100
# x["daaa"] Inf 6.59e+04 44582.73 10049.63 1207.337   100