Skip to content

Latest commit

 

History

History
370 lines (311 loc) · 10.7 KB

README.md

File metadata and controls

370 lines (311 loc) · 10.7 KB

validatedb

CRAN status R build status Codecov test coverage Mentioned in Awesome Official Statistics

validatedb executes validation checks written with R package validate on a database. This allows for checking the validity of records in a database.

Installation

You can install a development version with

remotes::install_github("data-cleaning/validatedb")

Example

library(validatedb)
#> Loading required package: validate

First we setup a table in a database (for demo purpose)

# create a table in a database
income <- data.frame(id=1:2, age=c(12,35), salary = c(1000,NA))
con <- DBI::dbConnect(RSQLite::SQLite())
DBI::dbWriteTable(con, "income", income)

We retrieve a reference/handle to the table in the DB with dplyr

tbl_income <- tbl(con, "income")
print(tbl_income)
#> # Source:   table<income> [?? x 3]
#> # Database: sqlite 3.37.2 []
#>      id   age salary
#>   <int> <dbl>  <dbl>
#> 1     1    12   1000
#> 2     2    35     NA

Let’s define a rule set and confront the table with it:

rules <- validator( is_adult   = age >= 18
                  , has_income = salary > 0
                  , mean_age   = mean(age,na.rm=TRUE) > 24
                  , has_values = is_complete(age, salary)
                  )

# and confront!
cf <- confront(tbl_income, rules, key = "id")

print(cf)
#> Object of class 'tbl_validation'
#> Call:
#>     confront.tbl_sql(tbl = dat, x = x, ref = ref, key = key, sparse = sparse)
#> 
#> Confrontations: 4
#> Tbl           : income ()
#> Key column    : id
#> Sparse        : FALSE
#> Fails         : [??] (see `values`, `summary`)
#> Errors        : 0

summary(cf)
#>                  name items npass nfail nNA warning error
#> is_adult     is_adult     2     1     1   0   FALSE FALSE
#> has_income has_income     2     1     0   1   FALSE FALSE
#> mean_age     mean_age     1     0     1   0   FALSE FALSE
#> has_values has_values     2     1     1   0   FALSE FALSE
#>                              expression
#> is_adult             age - 18 >= -1e-08
#> has_income                   salary > 0
#> mean_age   mean(age, na.rm = TRUE) > 24
#> has_values     is_complete(age, salary)

Values (i.e. validations on the table) can be retrieved like in validate with type="matrix" or type="list"

values(cf, type = "matrix")
#> [[1]]
#>      is_adult has_income has_values
#> [1,]    FALSE       TRUE       TRUE
#> [2,]     TRUE         NA      FALSE
#> 
#> [[2]]
#>      mean_age
#> [1,]    FALSE

But often this seems more handy:

values(cf, type = "tbl")
#> # Source:   lazy query [?? x 5]
#> # Database: sqlite 3.37.2 []
#>      id is_adult has_income mean_age has_values
#>   <int>    <int>      <int>    <int>      <int>
#> 1     1        0          1        0          1
#> 2     2        1         NA        0          0

or

values(cf, type = "tbl", sparse=TRUE)
#> # Source:   lazy query [?? x 3]
#> # Database: sqlite 3.37.2 []
#>      id rule        fail
#>   <int> <chr>      <int>
#> 1     1 is_adult       1
#> 2     2 has_income    NA
#> 3     1 mean_age       1
#> 4     2 mean_age       1
#> 5     2 has_values     1

We can see the sql code by using show_query:

show_query(cf)
#> <SQL>
#> SELECT `id`, CAST(`is_adult` AS BOOLEAN) AS `is_adult`, CAST(`has_income` AS BOOLEAN) AS `has_income`, CAST(`mean_age` AS BOOLEAN) AS `mean_age`, CAST(`has_values` AS BOOLEAN) AS `has_values`
#> FROM (SELECT `id`, NULLIF(`is_adult`, -1) AS `is_adult`, NULLIF(`has_income`, -1) AS `has_income`, NULLIF(`mean_age`, -1) AS `mean_age`, NULLIF(`has_values`, -1) AS `has_values`
#> FROM (SELECT `id`, MIN(`is_adult`) AS `is_adult`, MIN(`has_income`) AS `has_income`, MIN(`mean_age`) AS `mean_age`, MIN(`has_values`) AS `has_values`
#> FROM (SELECT `id`, CASE `rule` WHEN ('is_adult') THEN (COALESCE(1 - CAST(`fail` AS INTEGER), -1)) ELSE (1) END AS `is_adult`, CASE `rule` WHEN ('has_income') THEN (COALESCE(1 - CAST(`fail` AS INTEGER), -1)) ELSE (1) END AS `has_income`, CASE `rule` WHEN ('mean_age') THEN (COALESCE(1 - CAST(`fail` AS INTEGER), -1)) ELSE (1) END AS `mean_age`, CASE `rule` WHEN ('has_values') THEN (COALESCE(1 - CAST(`fail` AS INTEGER), -1)) ELSE (1) END AS `has_values`
#> FROM (SELECT `LHS`.`id` AS `id`, `rule`, `fail`
#> FROM (SELECT `id`
#> FROM `income`) AS `LHS`
#> LEFT JOIN (SELECT `id`, 'is_adult' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM `income`)
#> WHERE (`age` - 18.0 < -1e-08)
#> UNION ALL
#> SELECT `id`, 'is_adult' AS `rule`, NULL AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM `income`)
#> WHERE (((`age`) IS NULL))
#> UNION ALL
#> SELECT `id`, 'has_income' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `salary`
#> FROM `income`)
#> WHERE (`salary` <= 0.0)
#> UNION ALL
#> SELECT `id`, 'has_income' AS `rule`, NULL AS `fail`
#> FROM (SELECT `id`, `salary`
#> FROM `income`)
#> WHERE (((`salary`) IS NULL))
#> UNION ALL
#> SELECT `id`, 'mean_age' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM (SELECT `id`, `age`, AVG(`age`) OVER () AS `q01`
#> FROM (SELECT `id`, `age`
#> FROM `income`))
#> WHERE (`q01` <= 24.0))
#> UNION ALL
#> SELECT `id`, 'mean_age' AS `rule`, NULL AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM `income`)
#> WHERE (((`age`) IS NULL))
#> UNION ALL
#> SELECT `id`, 'has_values' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `age`, `salary`
#> FROM `income`)
#> WHERE (((`age`) IS NULL) OR ((`salary`) IS NULL))) AS `RHS`
#> ON (`LHS`.`id` = `RHS`.`id`)
#> ))
#> GROUP BY `id`))

Or write the sql to a file for documentation (and inspiration)

dump_sql(cf, "validation.sql")
------------------------------------------------------------
-- Do not edit, automatically generated with R package validatedb.
-- validatedb: 0.3.1.9000
-- validate: 1.1.0
-- R version 4.1.2 (2021-11-01)
-- Database: '', Table: 'income'
-- Date: 2022-03-14
------------------------------------------------------------

--------------------------------------
--  is_adult:  
--  validation rule:  age >= 18

SELECT `id`, 'is_adult' AS `rule`, 1 AS `fail`
FROM (SELECT `id`, `age`
FROM `income`)
WHERE (`age` - 18.0 < -1e-08)
UNION ALL
SELECT `id`, 'is_adult' AS `rule`, NULL AS `fail`
FROM (SELECT `id`, `age`
FROM `income`)
WHERE (((`age`) IS NULL))

--------------------------------------

UNION ALL

--------------------------------------
--  has_income:  
--  validation rule:  salary > 0

SELECT `id`, 'has_income' AS `rule`, 1 AS `fail`
FROM (SELECT `id`, `salary`
FROM `income`)
WHERE (`salary` <= 0.0)
UNION ALL
SELECT `id`, 'has_income' AS `rule`, NULL AS `fail`
FROM (SELECT `id`, `salary`
FROM `income`)
WHERE (((`salary`) IS NULL))

--------------------------------------

UNION ALL

--------------------------------------
--  mean_age:  
--  validation rule:  mean(age, na.rm = TRUE) > 24

SELECT `id`, 'mean_age' AS `rule`, 1 AS `fail`
FROM (SELECT `id`, `age`
FROM (SELECT `id`, `age`, AVG(`age`) OVER () AS `q01`
FROM (SELECT `id`, `age`
FROM `income`))
WHERE (`q01` <= 24.0))
UNION ALL
SELECT `id`, 'mean_age' AS `rule`, NULL AS `fail`
FROM (SELECT `id`, `age`
FROM `income`)
WHERE (((`age`) IS NULL))

--------------------------------------

UNION ALL

--------------------------------------
--  has_values:  
--  validation rule:  is_complete(age, salary)

SELECT `id`, 'has_values' AS `rule`, 1 AS `fail`
FROM (SELECT `id`, `age`, `salary`
FROM `income`)
WHERE (((`age`) IS NULL) OR ((`salary`) IS NULL))

--------------------------------------

Aggregate example

income <- data.frame(id = 1:2, age=c(12,35), salary = c(1000,NA))
con <- dbplyr::src_memdb()
tbl_income <- dplyr::copy_to(con, income, overwrite=TRUE)
print(tbl_income)
#> # Source:   table<income> [?? x 3]
#> # Database: sqlite 3.37.2 [:memory:]
#>      id   age salary
#>   <int> <dbl>  <dbl>
#> 1     1    12   1000
#> 2     2    35     NA

# Let's define a rule set and confront the table with it:
rules <- validator( is_adult   = age >= 18
                    , has_income = salary > 0
)

# and confront!
# in general with a db table it is handy to use a key
cf <- confront(tbl_income, rules, key="id")
aggregate(cf, by = "rule")
#> # Source:   lazy query [?? x 7]
#> # Database: sqlite 3.37.2 [:memory:]
#>   rule       npass nfail   nNA rel.pass rel.fail rel.NA
#>   <chr>      <int> <int> <int> <lgl>       <dbl>  <dbl>
#> 1 is_adult       1     1     0 NA            0.5    0  
#> 2 has_income     1     0     1 NA            0      0.5
aggregate(cf, by = "record")
#> # Source:   lazy query [?? x 3]
#> # Database: sqlite 3.37.2 [:memory:]
#>      id nfails   nNA
#>   <int>  <int> <int>
#> 1     1      1     0
#> 2     2      0     1

# to tweak performance of the db query the following options are available
# 1) store validation result in db
cf <- confront(tbl_income, rules, key="id", compute = TRUE)
# or identical
cf <- confront(tbl_income, rules, key="id")
cf <- compute(cf)

# 2) Store the validation sparsely
cf_sparse <- confront(tbl_income, rules, key="id", sparse=TRUE )

show_query(cf_sparse)
#> <SQL>
#> SELECT `id`, 'is_adult' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM `income`)
#> WHERE (`age` - 18.0 < -1e-08)
#> UNION ALL
#> SELECT `id`, 'is_adult' AS `rule`, NULL AS `fail`
#> FROM (SELECT `id`, `age`
#> FROM `income`)
#> WHERE (((`age`) IS NULL))
#> UNION ALL
#> SELECT `id`, 'has_income' AS `rule`, 1 AS `fail`
#> FROM (SELECT `id`, `salary`
#> FROM `income`)
#> WHERE (`salary` <= 0.0)
#> UNION ALL
#> SELECT `id`, 'has_income' AS `rule`, NULL AS `fail`
#> FROM (SELECT `id`, `salary`
#> FROM `income`)
#> WHERE (((`salary`) IS NULL))
values(cf_sparse, type="tbl")
#> # Source:   lazy query [?? x 3]
#> # Database: sqlite 3.37.2 [:memory:]
#>      id rule        fail
#>   <int> <chr>      <int>
#> 1     1 is_adult       1
#> 2     2 has_income    NA

validate specific functions

Added:

  • is_complete, all_complete
  • is_unique, all_unique
  • exists_any, exists_one
  • do_by, sum_by, mean_by, min_by, max_by

Todo

Some newly added validate utility functions are (still) missing from validatedb.

  • contains_exactly
  • is_linear_sequence
  • hierachy