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.
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")
# 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.
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:
- 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.
- 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'
andid == 0
: Data as a whole.var != '.all'
andid == 0
: Group (var
shouldn’t be an actual column name) or column (var
should be an actual column name) as a whole.var == '.all'
andid != 0
: Row as a whole.var != '.all'
andid != 0
: Described cell.
With exposure attached to data one can perform different kinds of actions: exploration, assertion, imputation and so on.
# 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
)
)
# 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")
)
# 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))
)
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)
)
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(.)))
)
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.
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 returnsTRUE
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.
More leaned towards assertions:
More leaned towards validation: