如何使功能执行得更快? [英] How to make function perform faster?

问题描述

我具有以下功能: http://i.stack.imgur.com/yXA67.png ，其中 mu 是矩阵(n_X行和n_Y列). d_X和d_Y是距离矩阵.

I have following function: http://i.stack.imgur.com/yXA67.png, where mu is matrix (n_X rows and n_Y columns). d_X and d_Y are distance matrices.

在R中实现此功能的一种方法是:

One way to implement this function in R would be:

H_mu <- function(mu, d_X, d_Y){

    value <- 0
    for(i in 1:nrow(d_X)){
       for(ii in 1:nrow(d_X)){
          for(j in 1:nrow(d_Y)){
             for(jj in 1:nrow(d_Y)){
                value <- value + mu[i,j]*mu[ii,jj]*abs(d_X[i,ii]-d_Y[j,jj])
          }}}} 
}

例如:

X <- matrix(rep(1,50),nrow = 50)
Y <- matrix(c(1:50),nrow = 50)
d_X <- as.matrix(dist(X, method = "euclidean", diag = T, upper = T))
d_Y <- as.matrix(dist(Y, method = "euclidean", diag = T, upper = T)) 
mu <- matrix(1/50, nrow = nrow(X), ncol = nrow(Y))

H_mu(mu, d_X, d_Y)
[1] 41650

> system.time(H_mu(mu, d_X, d_Y))
   user  system elapsed 
  22.67    0.01   23.06 

仅用50点计算就需要23秒.

Only with 50 points calculations take 23 seconds.

如何加快此功能?

推荐答案

似乎@Marat Talipov的建议是可行的.如果您不习惯使用C ++进行编码，则可以使用 typedFunction 自动生成用于简单的R函数.假定存在显式的return调用，它将使用R函数及其参数及其类型，并返回文本代码.

Seems like @Marat Talipov's suggestion is way to go. If you are not comfortable with coding in C++, you can use typedFunction to auto-generate Rcpp code for simple R functions. It takes R function and it's arguments along with their types, assuming that there is explicit return call, and returns text code.

 H_mu <- function(mu, d_X, d_Y){      
  value <- 0
  for(i in 1:nrow(d_X)){
    for(ii in 1:nrow(d_X)){
      for(j in 1:nrow(d_Y)){
        for(jj in 1:nrow(d_Y)){
          value <- value + mu[i,j]*mu[ii,jj]*abs(d_X[i,ii]-d_Y[j,jj])
        }}}} 
  return (value)
}

在这里，我已将return(value)添加到您的H_mu函数

Here I've added return(value) to your H_mu function

text <- typedFunction(H_mu, H_mu='double', value='double',
              mu='NumericVector',
              d_X='NumericVector',
              d_Y='NumericVector',
              i='int',
              ii='int',
              jj='int',
              j='int')
cat(text)

将结果复制并粘贴到Rcpp编辑器中，稍作调整后，您就可以执行H_mu_typed函数.

Copy-paste the outcome to your Rcpp editor, and after little tweaking you have executable H_mu_typed function.

Rcpp::cppFunction('double H_mu_typed(NumericMatrix mu, NumericMatrix d_X, NumericMatrix d_Y) {
  double value=0;
                  value = 0;
                  for (int i = 0; i <d_X.nrow(); i++) {
                  for (int ii = 0; ii < d_X.nrow(); ii++) {
                  for (int j = 0; j < d_Y.nrow(); j++) {
                  for (int jj = 0; jj < d_Y.nrow(); jj++) {
                  value = value + mu(i, j) * mu(ii, jj) * abs(d_X(i, ii) - d_Y(j, jj));
                  };
                  };
                  };
                  };
                  return(value);
                  }
                  ')

享受C ++的速度.

H_mu_typed(mu, d_X, d_Y)
[1] 41650

system.time(H_mu_typed(mu, d_X, d_Y))[3]
elapsed 
   0.01 

这篇关于如何使功能执行得更快?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！