Skip to content

echasnovski/ruler

Repository files navigation

ruler: Rule Your Data

Travis-CI Build Status R-CMD-check Coverage Status CRAN Dependencies Downloads

ruler offers a set of tools for creating tidy data validation reports using dplyr grammar of data manipulation. It is structured to be flexible and extendable in terms of creating rules and using their output.

To fully use this package a solid knowledge of dplyr is required. The key idea behind ruler’s design is to validate data by modifying regular dplyr code with as little overhead as possible.

Some functionality is powered by the keyholder package. It is highly recommended to use its supported functions during rule construction. All one- and two-table dplyr verbs applied to local data frames are supported and considered the most appropriate way to create rules.

This README is structured as follows:

  • Installation shows ways to install package.
  • Example shows the basic usage of ruler for exploration of obeying user-defined rules and its automatic validation.
  • Overview explains basic data and function types with design behind them.
  • Usage describes ruler’s capabilities in more detail.
  • Other packages for validation and assertions lists alternatives for described tasks.

Installation

You can install current stable version from CRAN with:

install.packages("ruler")

Also you can install development version from github with:

# install.packages("devtools")
devtools::install_github("echasnovski/ruler")

Example

# Utilities functions
is_integerish <- function(x) {
  all(x == as.integer(x))
}
z_score <- function(x) {
  abs(x - mean(x)) / sd(x)
}

# Define rule packs
my_packs <- list(
  data_packs(
    dims = . %>% summarise(nrow_low = nrow(.) >= 10, nrow_high = nrow(.) <= 15,
      ncol_low = ncol(.) >= 20, ncol_high = ncol(.) <= 30)
  ),
  group_packs(
    vs_am_num = . %>% group_by(vs, am) %>% summarise(vs_am_low = n() >= 7),
    .group_vars = c("vs", "am")
  ),
  col_packs(
    enough_col_sum = . %>%
      summarise_if(is_integerish, rules(is_enough = sum(.) >= 14))
  ),
  row_packs(
    enough_row_sum = . %>%
      filter(vs == 1) %>%
      transmute(is_enough = rowSums(.) >= 200)
  ),
  cell_packs(
    dbl_not_outlier = . %>%
      transmute_if(is.numeric, rules(is_not_out = z_score(.) < 1)) %>%
      slice(-(1:5))
  )
)

# Expose data to rules
mtcars_exposed <- mtcars %>% as_tibble() %>%
  expose(my_packs)

# View exposure
mtcars_exposed %>% get_exposure()
#>   Exposure
#> 
#> Packs info:
#> # A tibble: 5 × 4
#>   name            type       fun        remove_obeyers
#>   <chr>           <chr>      <list>     <lgl>         
#> 1 dims            data_pack  <data_pck> TRUE          
#> 2 vs_am_num       group_pack <grop_pck> TRUE          
#> 3 enough_col_sum  col_pack   <col_pack> TRUE          
#> 4 enough_row_sum  row_pack   <row_pack> TRUE          
#> 5 dbl_not_outlier cell_pack  <cell_pck> TRUE          
#> 
#> Tidy data validation report:
#> # A tibble: 117 × 5
#>   pack            rule       var      id value
#>   <chr>           <chr>      <chr> <int> <lgl>
#> 1 dims            nrow_high  .all      0 FALSE
#> 2 dims            ncol_low   .all      0 FALSE
#> 3 vs_am_num       vs_am_low  0.1       0 FALSE
#> 4 enough_col_sum  is_enough  am        0 FALSE
#> 5 enough_row_sum  is_enough  .all     19 FALSE
#> 6 dbl_not_outlier is_not_out mpg      15 FALSE
#> # ℹ 111 more rows

# Assert any breaker
invisible(mtcars_exposed %>% assert_any_breaker())
#>   Breakers report
#> Tidy data validation report:
#> # A tibble: 117 × 5
#>   pack            rule       var      id value
#>   <chr>           <chr>      <chr> <int> <lgl>
#> 1 dims            nrow_high  .all      0 FALSE
#> 2 dims            ncol_low   .all      0 FALSE
#> 3 vs_am_num       vs_am_low  0.1       0 FALSE
#> 4 enough_col_sum  is_enough  am        0 FALSE
#> 5 enough_row_sum  is_enough  .all     19 FALSE
#> 6 dbl_not_outlier is_not_out mpg      15 FALSE
#> # ℹ 111 more rows
#> Error: assert_any_breaker: Some breakers found in exposure.

Overview

Rule is a function which converts data unit of interest (data, group, column, row, cell) to logical value indicating whether this object satisfies certain condition.

Rule pack is a function which combines several rules into one functional block. The recommended way of creating rules is by creating packs right away with the use of dplyr and magrittr’s pipe operator.

Exposing data to rules means applying rules to data, collecting results in common format and attaching them to the data as an exposure attribute. In this way actual exposure can be done in multiple steps and also be a part of a general data preparation pipeline.

Exposure is a format designed to contain uniform information about validation of different data units. For reproducibility it also saves information about applied packs. Basically exposure is a list with two elements:

  1. Packs info: a tibble with the following structure:
    • name <chr> : Name of the pack. If not set manually it will be imputed during exposure.
    • type <chr> : Name of pack type. Indicates which data unit pack checks.
    • fun <list> : List of rule pack functions.
    • remove_obeyers <lgl> : Whether rows about obeyers (data units that obey certain rule) were removed from report after applying pack.
  2. Tidy data validation report: a tibble with the following structure:
    • pack <chr> : Name of rule pack from column ‘name’ in packs info.
    • rule <chr> : Name of the rule defined in rule pack.
    • var <chr> : Name of the variable which validation result is reported. Value ‘.all’ is reserved and interpreted as ‘all columns as a whole’. Note that var doesn’t always represent the actual column in data frame: for group packs it represents the created group name.
    • id <int> : Index of the row in tested data frame which validation result is reported. Value 0 is reserved and interpreted as ‘all rows as a whole’.
    • value <lgl> : Whether the described data unit obeys the rule.

There are four basic combinations of var and id values which define five basic data units:

  • var == '.all' and id == 0: Data as a whole.
  • var != '.all' and id == 0: Group (var shouldn’t be an actual column name) or column (var should be an actual column name) as a whole.
  • var == '.all' and id != 0: Row as a whole.
  • var != '.all' and id != 0: Described cell.

With exposure attached to data one can perform different kinds of actions: exploration, assertion, imputation and so on.

Usage

Creating packs

Data packs

# List of two rule packs for checking data properties
my_data_packs <- data_packs(
  # data_dims is a pack name
  data_dims = . %>% summarise(
    # ncol and nrow are rule names
    ncol = ncol(.) == 12,
    nrow = nrow(.) == 32
  ),

  # Data after subsetting should have number of rows in between 10 and 30
  # Rules are applied separately
  vs_1 = . %>% filter(vs == 1) %>%
    summarise(
      nrow_low = nrow(.) > 10,
      nrow_high = nrow(.) < 30
    )
)

Group packs

# List of one nameless rule pack for checking group property
my_group_packs <- group_packs(
  # Name will be imputed during exposure
  . %>% group_by(vs, am) %>%
    summarise(any_cyl_6 = any(cyl == 6)),

  # One should supply grouping variables for correct interpretation of output
  .group_vars = c("vs", "am")
)

Column packs

# rules() defines function predicators with necessary name imputations

# List of two rule pack for checking certain columns' properties
my_col_packs <- col_packs(
  sum_bounds = . %>% summarise_at(
    # Check only columns with names starting with 'c'
    vars(starts_with("c")),
    rules(sum_low = sum(.) > 300, sum_high = sum(.) < 400)
  ),

  # In the edge case of checking one column with one rule there is a need
  # for forcing inclusion of names in the output of summarise_at().
  # This is done with naming argument in vars()
  vs_mean = . %>% summarise_at(vars(vs = vs), rules(mean(.) > 0.5))
)

Row packs

z_score <- function(x) {
  (x - mean(x)) / sd(x)
}

# List of one rule pack checking certain rows' property
my_row_packs <- row_packs(
  row_mean = . %>% mutate(rowMean = rowMeans(.)) %>%
    transmute(is_common_row_mean = abs(z_score(rowMean)) < 1) %>%
    # Check only rows 10-15
    # Values in 'id' column of report will be based on input data (i.e. 10-15)
    # and not on output data (1-6)
    slice(10:15)
)

Cell packs

is_integerish <- function(x) {
  all(x == as.integer(x))
}

# List of two cell pack checking certain cells' property
my_cell_packs <- cell_packs(
  my_cell_pack_1 = . %>% transmute_if(
    # Check only integer-like columns
    is_integerish,
    rules(is_common = abs(z_score(.)) < 1)
  ) %>%
    # Check only rows 20-30
    slice(20:30),

  # The same edge case as in column rule pack
  vs_side = . %>% transmute_at(vars(vs = "vs"), rules(. > mean(.)))
)

Exposing

By default exposing removes obeyers.

mtcars %>%
  expose(my_data_packs, my_group_packs) %>%
  get_exposure()
#>   Exposure
#> 
#> Packs info:
#> # A tibble: 3 × 4
#>   name          type       fun        remove_obeyers
#>   <chr>         <chr>      <list>     <lgl>         
#> 1 data_dims     data_pack  <data_pck> TRUE          
#> 2 vs_1          data_pack  <data_pck> TRUE          
#> 3 group_pack__1 group_pack <grop_pck> TRUE          
#> 
#> Tidy data validation report:
#> # A tibble: 3 × 5
#>   pack          rule      var      id value
#>   <chr>         <chr>     <chr> <int> <lgl>
#> 1 data_dims     ncol      .all      0 FALSE
#> 2 group_pack__1 any_cyl_6 0.0       0 FALSE
#> 3 group_pack__1 any_cyl_6 1.1       0 FALSE

One can leave obeyers by setting .remove_obeyers to FALSE.

mtcars %>%
  expose(my_data_packs, my_group_packs, .remove_obeyers = FALSE) %>%
  get_exposure()
#>   Exposure
#> 
#> Packs info:
#> # A tibble: 3 × 4
#>   name          type       fun        remove_obeyers
#>   <chr>         <chr>      <list>     <lgl>         
#> 1 data_dims     data_pack  <data_pck> FALSE         
#> 2 vs_1          data_pack  <data_pck> FALSE         
#> 3 group_pack__1 group_pack <grop_pck> FALSE         
#> 
#> Tidy data validation report:
#> # A tibble: 8 × 5
#>   pack          rule      var      id value
#>   <chr>         <chr>     <chr> <int> <lgl>
#> 1 data_dims     ncol      .all      0 FALSE
#> 2 data_dims     nrow      .all      0 TRUE 
#> 3 vs_1          nrow_low  .all      0 TRUE 
#> 4 vs_1          nrow_high .all      0 TRUE 
#> 5 group_pack__1 any_cyl_6 0.0       0 FALSE
#> 6 group_pack__1 any_cyl_6 0.1       0 TRUE 
#> # ℹ 2 more rows

By default expose() guesses the pack type if ‘not-pack’ function is supplied. This behaviour has some edge cases but is useful for interactive use.

mtcars %>%
  expose(
    some_data_pack = . %>% summarise(nrow = nrow(.) == 10),
    some_col_pack = . %>% summarise_at(vars(vs = "vs"), rules(is.character(.)))
  ) %>%
  get_exposure()
#>   Exposure
#> 
#> Packs info:
#> # A tibble: 2 × 4
#>   name           type      fun        remove_obeyers
#>   <chr>          <chr>     <list>     <lgl>         
#> 1 some_data_pack data_pack <data_pck> TRUE          
#> 2 some_col_pack  col_pack  <col_pack> TRUE          
#> 
#> Tidy data validation report:
#> # A tibble: 2 × 5
#>   pack           rule    var      id value
#>   <chr>          <chr>   <chr> <int> <lgl>
#> 1 some_data_pack nrow    .all      0 FALSE
#> 2 some_col_pack  rule__1 vs        0 FALSE

To write strict and robust code one can set .guess to FALSE.

mtcars %>%
  expose(
    some_data_pack = . %>% summarise(nrow = nrow(.) == 10),
    some_col_pack = . %>% summarise_at(vars(vs = "vs"), rules(is.character(.))),
    .guess = FALSE
  ) %>%
  get_exposure()
#> Error in expose_single.default(X[[i]], ...): There is unsupported class of rule pack.

Acting after exposure

General actions are recommended to be done with act_after_exposure(). It takes two arguments:

  • .trigger - a function which takes the data with attached exposure and returns TRUE if some action should be made.
  • .actor - a function which takes the same argument as .trigger and performs some action.

If trigger didn’t notify then the input data is returned untouched. Otherwise the output of .actor() is returned. Note that act_after_exposure() is often used for creating side effects (printing, throwing error etc.) and in that case should invisibly return its input (to be able to use it with pipe).

trigger_one_pack <- function(.tbl) {
  packs_number <- .tbl %>%
    get_packs_info() %>%
    nrow()

  packs_number > 1
}

actor_one_pack <- function(.tbl) {
  cat("More than one pack was applied.\n")

  invisible(.tbl)
}

mtcars %>%
  expose(my_col_packs, my_row_packs) %>%
  act_after_exposure(
    .trigger = trigger_one_pack,
    .actor = actor_one_pack
  ) %>%
  invisible()
#> More than one pack was applied.

ruler has function assert_any_breaker() which can notify about presence of any breaker in exposure.

mtcars %>%
  expose(my_col_packs, my_row_packs) %>%
  assert_any_breaker()
#>   Breakers report
#> Tidy data validation report:
#> # A tibble: 4 × 5
#>   pack       rule               var      id value
#>   <chr>      <chr>              <chr> <int> <lgl>
#> 1 sum_bounds sum_low            cyl       0 FALSE
#> 2 sum_bounds sum_low            carb      0 FALSE
#> 3 vs_mean    rule__1            vs        0 FALSE
#> 4 row_mean   is_common_row_mean .all     15 FALSE
#> Error: assert_any_breaker: Some breakers found in exposure.

Other packages for validation and assertions

More leaned towards assertions:

More leaned towards validation: