Skip to content

A system for stashing and loading the results of long running computations.

License

Notifications You must be signed in to change notification settings

jhrcook/mustashe

Repository files navigation

mustashe

CRAN status CRAN downloads R-CMD-check Codecov test coverage License: GPL v3

The goal of ‘mustashe’ is to save time on long-running computations by storing and reloading the resulting object after the first run. The next time the computation is run, instead of evaluating the code, the stashed object is loaded. ‘mustashe’ is great for storing intermediate objects in an analysis.

Installation

You can install the released version of ‘mustashe’ from CRAN with:

install.packages("mustashe")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("jhrcook/mustashe")

Loading ‘mustashe’

The ‘mustashe’ package is loaded like any other, using the library() function.

library(mustashe)

Basic example

Below is a simple example of how to use the stash() function from ‘mustashe’.

Let’s say, for part of an analysis, we are running a long simulation to generate random data rnd_vals. This is mocked below using the Sys.sleep() function. We can time this process using the ‘tictoc’ library.

tictoc::tic("random simulation")
stash("rnd_vals", {
  Sys.sleep(3)
  rnd_vals <- rnorm(1e5)
})
#> Stashing object.
tictoc::toc()
#> random simulation: 3.044 sec elapsed

Now, if we come back tomorrow and continue working on the same analysis, the second time this process is run the code is not evaluated because the code passed to stash() has not changed. Instead, the random values rnd_vals is loaded.

tictoc::tic("random simulation")
stash("rnd_vals", {
  Sys.sleep(3)
  rnd_vals <- rnorm(1e5)
})
#> Loading stashed object.
tictoc::toc()
#> random simulation: 0.023 sec elapsed

Dependencies

A common problem with storing intermediates is that they have dependencies that can change. If a dependency changes, then we want the stashed value to be updated. This is accomplished by passing the names of the dependencies to the depends_on argument.

For instance, let’s say we are calculating some value foo using x. (For the following example, I will use a print statement to indicate when the code is evaluated.)

x <- 100

stash("foo", depends_on = "x", {
  print("Calculating `foo` using `x`.")
  foo <- x + 1
})
#> Stashing object.
#> [1] "Calculating `foo` using `x`."

foo
#> [1] 101

Now if x is not changed, then the code for foo does not get re-evaluated.

x <- 100

stash("foo", depends_on = "x", {
  print("Calculating `foo` using `x`.")
  foo <- x + 1
})
#> Loading stashed object.

foo
#> [1] 101

But if x does change, then foo gets re-evaluated.

x <- 200

stash("foo", depends_on = "x", {
  print("Calculating `foo` using `x`.")
  foo <- x + 1
})
#> Updating stash.
#> [1] "Calculating `foo` using `x`."

foo
#> [1] 201

Other API features

Functional interface

In the examples above, stash() does not return a value (actually, it invisibly returns NULL), instead assigning the result of the computation to an object named using the var argument. Frequently, though, a return value is desired. This behavior can be induced by setting the argument functional = TRUE.

b <- stash("b", functional = FALSE, {
  rnorm(5, 0, 1)
})
#> Stashing object.
b
#> NULL
b <- stash("b", functional = TRUE, {
  rnorm(5, 0, 1)
})
#> Loading stashed object.
b
#> [1]  0.26499342  1.83074748 -0.05937826 -0.05320937  0.43790418

Functions as dependencies

The stash() function can take other functions as dependencies. The body and formals components of the function object are checked to see if they have changed. (More information on the structure of function objects in R can be found in Hadley Wickham’s Advanced R - Functions: Function components.)

As an example, suppose you have a script with the following code. It is run, and the value of 5 is stashed for a and it is dependent on the function add_x().

add_x <- function(y, x = 2) {
  y + x
}

stash("a", depends_on = "add_x", {
  a <- add_x(3)
})
#> Stashing object.
a
#> [1] 5

You continue working and change the function add_x() to use the default value of 5 instead of 2. This change will cause the code for a to be re-run and a will be assigned the value 8. Note that the code in the code argument for stash() did not change, the code was re-run because a dependency changed.

add_x <- function(y, x = 5) {
  y + x
}

stash("a", depends_on = "add_x", {
  a <- add_x(3)
})
#> Updating stash.
a
#> [1] 8

Using stash() in functions

Because of the careful management of R environments, stash() can be used inside of functions. In the example below, note that the stashed object will depend on the value of the magic_number object in the function.

magic_number <- 10
do_data_science <- function() {
  magic_number <- 5
  stash("rand_num", depends_on = c("magic_number"), {
    runif(1, 0, 10)
  })
  return(rand_num)
}

do_data_science()
#> Stashing object.
#> [1] 9.094619

Changing the value of the magic_number object in the global environment will not invalidate the stash.

magic_number <- 11
do_data_science()
#> Loading stashed object.
#> [1] 9.094619

Stashing results of sourcing a R script

It is also possible to stash the results of sourcing and R script. The contents of the script are an implicit dependency for the stash, so if the script changes, it will be re-sourced the next time around. It is also possible to include additional dependencies using the depends_on parameter in the same way as with a regular stash.

The natural behavior of the source() function is maintained by returning the last evaluated value in the script.

# Write a temporary R script.
temp_script <- tempfile()
write("print('Script to get 5 letters'); sample(letters, 5)", temp_script)

x <- stash_script(temp_script)
#> Stashing object.
#> [1] "Script to get 5 letters"
x
#> [1] "d" "t" "l" "o" "u"
x2 <- stash_script(temp_script)
#> Loading stashed object.
x2
#> [1] "d" "t" "l" "o" "u"

Configuration

Using ‘here’ to create file paths

The ‘here’ package is useful for handling file paths in R projects, particularly when using an RStudio project. The main function, here::here(), can be used to create the file path for stashing an object by setting the ‘mustashe’ configuration option with the config_mustashe() function.

config_mustashe(use_here = TRUE)

This behavior can be turned off, too.

config_mustashe(use_here = FALSE)

Other options

Defaults for the verbose and functional (see above) arguments of stashing functions can also be configured. For example, you can have the functions run silently and return the result by default.

config_mustashe(verbose = FALSE, functional = TRUE)

Acknowledgements

Contributors

I would like to thank the contributors to this package for their additions of key features and bug squashing:

  • vinayakvsv fixed an annoying bug early on in the development of the library.
  • jimbrig replaced the file read/write system with the ‘qs’ library.
  • traversc introduced the functional API to stash().
  • torfason upgraded R environment management enabling stashing in functions and linking functions as dependencies to a stashed object. He also created stash_script().

Attribution

The inspiration for this package came from the cache() feature in the ‘ProjectTemplate’ package. While the functionality and implementation are a bit different, this would have been far more difficult to do without referencing the source code from ‘ProjectTemplate’.


Contact

Any issues and feedback on ‘mustashe’ can be submitted here. Alternatively, I can be reached through the contact form on my website or on Twitter @JoshDoesa