readd() returns an object from the cache, and loadd() loads one or more objects from the cache into your environment or session. These objects are usually targets built by make(). If target is dynamic, readd() and loadd() retrieve a list of sub-target values. You can restrict which sub-targets to include using the subtargets argument.

  character_only = FALSE,
  path = NULL,
  search = NULL,
  cache = drake::drake_cache(path = path),
  namespace = NULL,
  verbose = 1L,
  show_source = FALSE,
  subtargets = NULL,
  subtarget_list = FALSE

  list = character(0),
  imported_only = NULL,
  path = NULL,
  search = NULL,
  cache = drake::drake_cache(path = path),
  namespace = NULL,
  envir = parent.frame(),
  jobs = 1,
  verbose = 1L,
  deps = FALSE,
  lazy = "eager",
  graph = NULL,
  replace = TRUE,
  show_source = FALSE,
  tidyselect = !deps,
  config = NULL,
  subtargets = NULL,
  subtarget_list = FALSE



If character_only is TRUE, then target is a character string naming the object to read. Otherwise, target is an unquoted symbol with the name of the object.


Logical, whether name should be treated as a character or a symbol (just like character.only in library()).


Path to a drake cache (usually a hidden .drake/ folder) or NULL.




drake cache. See new_cache(). If supplied, path is ignored.


Optional character string, name of the storr namespace to read from.


Deprecated on 2019-09-11.


Logical, option to show the command that produced the target or indicate that the object was imported (using show_source()).


A numeric vector of indices. If target is dynamic, loadd() and readd() retrieve a list of sub-targets. You can restrict which sub-targets to retrieve with the subtargets argument. For example, readd(x, subtargets = seq_len(3)) only retrieves the first 3 sub-targets of dynamic target x.


Logical, for dynamic targets only. If TRUE, the dynamic target is loaded as a named list of sub-target values. If FALSE, drake attempts to concatenate the sub-targets with vctrs::vec_c() (and returns an unnamed list if such concatenation is not possible).


Targets to load from the cache: as names (symbols) or character strings. If the tidyselect package is installed, you can also supply dplyr-style tidyselect commands such as starts_with(), ends_with(), and one_of().


Character vector naming targets to be loaded from the cache. Similar to the list argument of remove().


Logical, deprecated.


Environment to load objects into. Defaults to the calling environment (current workspace).


Number of parallel jobs for loading objects. On non-Windows systems, the loading process for multiple objects can be lightly parallelized via parallel::mclapply(). just set jobs to be an integer greater than 1. On Windows, jobs is automatically demoted to 1.


Logical, whether to load any cached dependencies of the targets instead of the targets themselves.

Important note: deps = TRUE disables tidyselect functionality. For example, loadd(starts_with("model_"), config = config, deps = TRUE) does not work. For the selection mechanism to work, the model_* targets to need to already be in the cache, which is not always the case when you are debugging your projects. To help drake understand what you mean, you must name the targets explicitly when deps is TRUE, e.g. loadd(model_A, model_B, config = config, deps = TRUE).


Either a string or a logical. Choices:

  • "eager": no lazy loading. The target is loaded right away with assign().

  • "promise": lazy loading with delayedAssign()

  • "bind": lazy loading with active bindings: bindr::populate_env().

  • TRUE: same as "promise".

  • FALSE: same as "eager".




Logical. If FALSE, items already in your environment will not be replaced.


Logical, whether to enable tidyselect expressions in ... like starts_with("prefix") and ends_with("suffix").


Optional drake_config() object. You should supply one if deps is TRUE.


The cached value of the target.


There are three uses for the loadd() and readd() functions:

  1. Exploring the results outside the drake/make() pipeline. When you call make() to run your project, drake puts the targets in a cache, usually a folder called .drake. You may want to inspect the targets afterwards, possibly in an interactive R session. However, the files in the .drake folder are organized in a special format created by the storr package, which is not exactly human-readable. To retrieve a target for manual viewing, use readd(). To load one or more targets into your session, use loadd().

  2. In knitr / R Markdown reports. You can borrow drake targets in your active code chunks if you have the right calls to loadd() and readd(). These reports can either run outside the drake pipeline, or better yet, as part of the pipeline itself. If you call knitr_in("your_report.Rmd") inside a drake_plan() command, then make() will scan "your_report.Rmd" for calls to loadd() and readd() in active code chunks, and then treat those loaded targets as dependencies. That way, make() will automatically (re)run the report if those dependencies change.

  3. If you are using make(memory_strategy = "none") or make(memory_strategy = "unload"), loadd() and readd() can manually load dependencies into memory for the target that is being built. If you do this, you must carefully inspect deps_target() and vis_drake_graph() before running make() to be sure the dependency relationships among targets are correct. If you do not wish to incur extra dependencies with loadd() or readd(), you will need to use ignore(), e.g. drake_plan(x = 1, y = ignore(readd(x))) or drake_plan(x = 1, y = readd(ignore("x"), character_only = TRUE)). Compare those plans to drake_plan(x = 1, y = readd(x)) and drake_plan(x = 1, y = readd("x", character_only = TRUE)) using vis_drake_graph() and deps_target().

See also


if (FALSE) { isolate_example("Quarantine side effects.", { if (suppressWarnings(require("knitr"))) { load_mtcars_example() # Get the code with drake_example("mtcars"). make(my_plan) # Run the project, build the targets. readd(reg1) # Return imported object 'reg1' from the cache. readd(small) # Return targets 'small' from the cache. readd("large", character_only = TRUE) # Return 'large' from the cache. # For external files, only the fingerprint/hash is stored. readd(file_store(""), character_only = TRUE) } }) } if (FALSE) { isolate_example("Quarantine side effects.", { if (suppressWarnings(require("knitr"))) { load_mtcars_example() # Get the code with drake_example("mtcars"). make(my_plan) # Run the projects, build the targets. config <- drake_config(my_plan) loadd(small) # Load target 'small' into your workspace. small # For many targets, you can parallelize loadd() # using the 'jobs' argument. loadd(list = c("small", "large"), jobs = 2) ls() # Load the dependencies of the target, coef_regression2_small loadd(coef_regression2_small, deps = TRUE, config = config) ls() # Load all the targets listed in the workflow plan # of the previous `make()`. # If you do not supply any target names, `loadd()` loads all the targets. # Be sure your computer has enough memory. loadd() ls() } }) }