plyr なんて...（その9）

ある分析を何回か繰り返し，その結果を保存しておくという例

> f = function() {lm(Ozone ~ Temp, aq[sample(nrow(aq), 100, replace=TRUE),])}
> f1 = function() rlply(100, f)
> set.seed(1); ans1 = f1(); head(ans1, 2)
[[1]]

Call:
lm(formula = Ozone ~ Temp, data = aq[sample(nrow(aq), 100, replace = TRUE),
    ])

Coefficients:
(Intercept)         Temp  
   -153.799        2.559  


[[2]]

Call:
lm(formula = Ozone ~ Temp, data = aq[sample(nrow(aq), 100, replace = TRUE),
    ])

Coefficients:
(Intercept)         Temp  
   -147.962        2.418  


>
> f2 = function() lapply(1:100, function(i) f())
> set.seed(1); ans2 = f2(); head(ans2, 2)
[[1]]

Call:
lm(formula = Ozone ~ Temp, data = aq[sample(nrow(aq), 100, replace = TRUE),
    ])

Coefficients:
(Intercept)         Temp  
   -153.799        2.559  


[[2]]

Call:
lm(formula = Ozone ~ Temp, data = aq[sample(nrow(aq), 100, replace = TRUE),
    ])

Coefficients:
(Intercept)         Temp  
   -147.962        2.418  


>
> # lm の結果が全部がほしいわけではないというならば以下のように
> # rlply の定義の主要部は replicate とほとんど同じ
> f3 = function() replicate(100, f()$coefficients[2])
> set.seed(1); ans3 = f3(); head(ans3, 2)
    Temp     Temp
2.559170 2.418056
>
> # 結果的にも，引き分け（replicate が若干速い？）
> benchmark(f1(), f2(), f3(), columns=c("test", "replications", "elapsed", "relative", "user.self", "sys.self"), replications=100, order=NULL)
  test replications elapsed relative user.self sys.self
1 f1()          100  15.485    1.063    15.488    0.120
2 f2()          100  15.295    1.050    15.298    0.090
3 f3()          100  14.572    1.000    14.582    0.078