2012-01-08 76 views
4

我怎样才能加速以下循环?R加速while循环

count <- function(start, stepsize, threshold) { 
    i <- 1; 
    while (start <= threshold) { 
    start <- stepsize*i+start; 
    i <- i+1; 
    } 
    return(i-1); 
} 

system.time(count(1, 0.004, 1e10)) 
+1

对于这个特定的问题,我会制定出手工加总 - 例如你知道从1到n的和(i)是n *(n + 1) - 然后求解适当的二次方程并进行调整。你也可以字节编译......是更大问题的一部分,还是只需要解决这个确切的问题? – 2012-01-08 15:17:19

+0

就是这样!非常感谢! – rua 2012-01-08 16:27:28

回答

10

工作出款项在上面的评论:

## start + S*n*(n-1)/2 = T 
## (T-start)*2/S = n*(n-1) 
## n*(n-1) - (T-start)*2/S = 0 

的函数来解决这个二次方程:

ff <- function(start,stepsize,threshold) { 
    C <- (threshold-start)*2/stepsize 
    ceiling((-1 + sqrt(1+4*C))/2) 
} 

该解决方案基本上不花时间......

> system.time(cc <- count(1, 0.004, 1e10)) 
    user system elapsed 
    5.372 0.056 5.642 
> system.time(cc2 <- ff(1, 0.004, 1e10)) 
    user system elapsed 
     0  0  0 
> cc2 
[1] 2236068 
> cc 
[1] 2236068 

该曲关键是这是否能够推广到您需要解决的确切问题。

0

它看起来像你试图做到这一点:

recount <- function(start, stepsize, threshold) { 
    NewCount <<- floor((threshold-start)/stepsize) 
} 
(fast <- system.time(recount(1, 0.004, 1e10))) 

它不带任何可测量的时间。

没有全局变量,这里是什么样子:

recount <- function(start, stepsize, threshold) { 
    return(floor((threshold-start)/stepsize)) 
} 
(fast <- system.time(NewCount <- recount(1, 0.004, 1e10))) 
+1

??但是这并没有给出与OP的测试代码相同的答案......并且设置一个全局变量而不是返回一个值是相当单一的... – 2012-01-12 16:13:44

0

有一个有趣的博客如何加快R中循环使用的一些技巧

Another aspect of speeding up loops in R

这是例如在该页面报告

NROW=5000 
NCOL=100 

#Ex. 1 - Creation of a results matrix where its memory 
#allocation must continually be redefined 
t1 <- Sys.time() 
x <- c() 
for(i in seq(NROW)){ 
    x <- rbind(x, runif(NCOL)) 
} 
T1 <- Sys.time() - t1 


#Ex. 2 - Creation of a results matrix where its memory 
#allocation is defined only once, at the beginning of the loop. 
t2 <- Sys.time() 
x <- matrix(NA, nrow=NROW, ncol=NCOL) 
for(i in seq(NROW)){ 
    x[i,] <- runif(NCOL) 
} 
T2 <- Sys.time() - t2 


#Ex. 3 - Creation of a results object as an empty list of length NROW. 
#Much faster than Ex. 1 even though the size of the list is 
#not known at the start of the loop. 
t3 <- Sys.time() 
x <- vector(mode="list", NROW) 
for(i in seq(NROW)){ 
    x[[i]] <- runif(NCOL) 
} 
T3 <- Sys.time() - t3 

png("speeding_up_loops.png") 
barplot(c(T1, T2, T3), names.arg = c("Concatenate result", "Fill empty matrix", "Fill empty list"),ylab="Time in seconds") 
dev.off() 

T1;T2;T3