Skip to content

Latest commit

 

History

History
124 lines (102 loc) · 4.04 KB

average-dataset-by-R-buitin-functions.md

File metadata and controls

124 lines (102 loc) · 4.04 KB
Raw

Different ways and time required to calculate mean of a dataset with R buit-in functions

Let (x_i = (x_{i1}, x_{i2}, ..., x_{id}), 1 \leq i \leq n = 150) (nrow(iris)), (1 \leq d \leq 4) (ncol(iris)) be each object of the iris dataset. The mean of the dataset is defined as

[ \frac{\sum_{i = 1}^n x_i}{n} = \frac{\sum_{i = 1}^n(x_{i1}, x_{i2}, x_{i3}, x_{i4})}{n} ]

mean.by.column = function(X){
    v.mean = c()
    for (i in 1:ncol(X)){
        v.mean[i] = mean(X[, i])
    }
    names(v.mean) = names(X)
    v.mean
}
mean.by.sum.nrow = function(X){
    v.mean = c()
    for (i in 1:ncol(X)){
        v.mean[i] = sum(X[, i])/nrow(X)
    }
    names(v.mean) = names(X)
    v.mean
}

Time required to compute each function

Using the rbenchmark package

library(rbenchmark)
time_required = function(X){
    out = list()
    time = benchmark("colMeans" = {
            out$colmeans = colMeans(X)
          },
          "colSums/nrow" = {
            out$colsums = colSums(X)/nrow(X)
          },
          "apply" = {
            out$apply = apply(X, 2, mean)
          },
          "mean.by.column" = {
            out$mean1 = mean.by.column(X)
          },
          "mean.by.sum.nrow" = {
            out$mean2 = mean.by.sum.nrow(X)
          },
          replications = 1000,
          columns = c("test", "replications", "elapsed",
                      "relative", "user.self", "sys.self"))
    out$time = time[order(time$elapsed), ]
    out
}

Mean of the iris dataset

result = time_required(iris[, -5])

knitr::kable(result$time)
test replications elapsed relative user.self sys.self
5 mean.by.sum.nrow 1000 0.04 1.00 0.05 0
2 colSums/nrow 1000 0.05 1.25 0.04 0
4 mean.by.column 1000 0.05 1.25 0.04 0
1 colMeans 1000 0.08 2.00 0.08 0
3 apply 1000 0.11 2.75 0.11 0

Mean of the wine dataset

wine.uci = read.csv("https://archive.ics.uci.edu/ml/machine-learning-databases/wine/wine.data", header = FALSE)
knitr::kable(time_required(wine.uci[, -1])$time)
test replications elapsed relative user.self sys.self
2 colSums/nrow 1000 0.09 1.000 0.09 0.00
1 colMeans 1000 0.13 1.444 0.10 0.01
5 mean.by.sum.nrow 1000 0.15 1.667 0.16 0.00
4 mean.by.column 1000 0.17 1.889 0.17 0.00
3 apply 1000 0.26 2.889 0.27 0.00

Mean of the breast cancer dataset

breast.cancer.uci = read.csv("https://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/wdbc.data", header = FALSE)
knitr::kable(time_required(breast.cancer.uci[,-2])$time)
test replications elapsed relative user.self sys.self
2 colSums/nrow 1000 0.22 1.000 0.22 0
1 colMeans 1000 0.25 1.136 0.25 0
5 mean.by.sum.nrow 1000 0.34 1.545 0.34 0
4 mean.by.column 1000 0.38 1.727 0.38 0
3 apply 1000 0.69 3.136 0.69 0

Considerations:

  • The function mean.by.column that calculates the mean of each column inside a for was faster than using the apply function.
  • The functions mean.by.sum.nrow and colSums/nrow were faster than the colMeans function.