README.md
In benchr: High Precise Measurement of R Expressions Execution Time

benchr

Package benchr provides an infrastructure (framework) for precise measurement of R expressions execution time.

Cross-platform implementation of the timer (the same code for all supported platforms);
High precision measurement of time intervals: usually nano or microseconds;
The reliability of the results due to a preliminary estimation of the timer error and subsequent correction of measurement results;
The stability of the results due to multiple repetitions of the measurements and the use of robust (resistant to outliers) statistics (quantile);
Informative output, including measurement accuracy, execution regime and descriptive statistics for each expression;
Various graphical representation of measurement results, including box plots, scatter plots and violin plots.

To install the package from the CRAN run the following command:

install.packages("benchr", repos = "https://cloud.r-project.org/")

To install the development version the following command can be used:

install.packages("benchr", repos = "https://artemklevtsov.gitlab.io/benchr")

This package contains the compiled code, so to install it on Windows you will also need Rtools.

To measure execution time of arbitrary R code, benchr provides function benchmark(), as well as a number of additional methods for analysis and representation of results. Here’s an example of time measurement for several expressions.

library(benchr)
benchmark(rep(1:100, each = 10), rep.int(1:100, rep.int(10, 100)))
#> Benchmark summary:
#> Time units : microseconds 
#>                              expr n.eval  min lw.qu median  mean up.qu  max total relative
#>             rep(1:100, each = 10)    100 20.7 21.10  21.20 22.60 21.40 48.3  2260     2.97
#>  rep.int(1:100, rep.int(10, 100))    100  6.4  6.89   7.16  7.52  7.43 27.4   752     1.00
identical(rep(1:100, each = 10), rep.int(1:100, rep.int(10, 100)))
#> [1] TRUE

The resulting object can be saved as a variable and reused later in further methods:

res <- benchmark(NULL, {NULL}, {{{NULL}}})
summary(res)
#> Time units : nanoseconds 
#>              expr n.eval min lw.qu median  mean up.qu   max total relative
#>              NULL    100   4     7   13.0  11.4    14    50  1140     1.00
#>          { NULL }    100  38    43   48.5  58.6    55   850  5860     3.73
#>  { { { NULL } } }    100 118   122  129.0 428.0   138 29100 42800     9.92

To present the results of measurements implemented additional methods for the class benchmark object:

mean – means and confidence intervals for each R expression;
summary – statistics (quantiles, means) for each R expression;
print – text representation of results based on method summary;
plot – scatter plot the execution time of each expression measure;
boxplot – box plot the execution time of each expression.

For further details refer to the manual pages and vignettes:

help(package = "benchr")

Use the following command to go to the page for bug report submissions:

bug.report(package = "benchr")

Before reporting a bug or submitting an issue, please do the following:

Make sure that no error was found and corrected previously identified. You can use the search by the bug tracker;
Check the news list for the current version of the package. An error it might have been caused by changes in the package. This can be done with news(package = "benchr", Version == packageVersion("benchr")) command;
Make a minimal reproducible example of the code that consistently causes the error;
Make sure that the error triggered in the function from the benchr package, rather than in the code that you pass, that is other functions or packages;
Try to reproduce the error with the last development version of the package from the git repository.

When submitting a bug report please include the output produced by functions traceback() and sessionInfo(). This may save a lot of time.

The benchr package is distributed under GPLv2 license.