dcmodifydb

Documented reproducible data correction on a database

Introduction

The goal of dcmodifydb is to apply modification rules specified with dcmodify on a database table, allowing for documented, reproducable data cleaning adjustments in a database.

dcmodify separates intent from execution: a user specifies the what , why and how of an automatic data change and uses dcmodifydb to execute them on a tbl database table.

dcmodidfydb is optimized and restricted to database tables that can be accessed within R through DBI. It uses the dbplyr package to translate the data correction rules in R syntax into SQL syntax. The advantage of this approach is that all data correction is done within the database, which may be a requirement of your organisation or because the data table is simply too large to be held in memory. A disadvantage is that not all R statements can be translated into SQL statement, so dcmodifydb is more restricted than dcmodify which can use the full R potential. Nonetheless dcmodifydb may be sufficient and efficient for many use cases.

For common error scenario’s see vignette("scenarios", package="dcmodifydb"). For the supported syntax for specifying rules see vignette("syntax", package="dcmodifydb").

Installation

dcmodifydb can be installed with

install.packages("dcmodifydb")

and loaded with:

library(dcmodify)
library(dcmodifydb)

Usage

dcmodifydb works on a database table, so we need a connection to a table within a database.

We set up a database table with sqlite using the person data set, but for your use case you should connect to your database.

income age gender year smokes cigarettes
2000 12 M 2020 no 10
2010 14 f 2019 yes 4
2010 25 v 19 no NA
1010 65 M 20 yes NA
con <- DBI::dbConnect(RSQLite::SQLite(), ":memory:")
dplyr::copy_to(con, person)

We now retrieve a handle to the person table using dplyr

person_tbl <- dplyr::tbl(con, "person")
person_tbl
#> # Source:   table<person> [4 x 6]
#> # Database: sqlite 3.38.5 [:memory:]
#>   income   age gender  year smokes cigarettes
#>    <int> <int> <chr>  <int> <chr>       <int>
#> 1   2000    12 M       2020 no             10
#> 2   2010    14 f       2019 yes             4
#> 3   2010    25 v         19 no             NA
#> 4   1010    65 M         20 yes            NA

The person dataset clearly contains some errors that can be corrected. We specify the corrections using modifier rules and apply them directly with the functionmodify_so.

First we correct that children can not have an income and that year must be a long year.

library(dcmodify) # needed for modifying rules
library(dcmodifydb) # needed to translate the rules
modify_so( person_tbl
         , if (age < 16) income = 0
         , if (year < 25) year = year + 2000
         )
#> Warning: `copy` not specified, setting `copy=TRUE`, working on copy of table.
#> # Source:   table<dcmodifydb_8543374> [4 x 6]
#> # Database: sqlite 3.38.5 [:memory:]
#>   income   age gender  year smokes cigarettes
#>    <int> <int> <chr>  <int> <chr>       <int>
#> 1      0    12 M       2020 no             10
#> 2      0    14 f       2019 yes             4
#> 3   2010    25 v       2019 no             NA
#> 4   1010    65 M       2020 yes            NA

Note that the corrections are made on a copy of the table by default, to avoid accidents with the data.

A better approach than directly applying corrections is to store the rules in a modifier object and apply them in a separate step to a data base table.

This makes it easier to maintain, use and document a set of rules. With dcmodify one can specify rules with the function modifier:

# separate rule set
m <- modifier( if (age < 16) income = 0
             , if (year < 25) year = year + 2000
             , if (cigarettes > 0 ) smokes = "yes"
             , if (smokes == "no") cigarettes = 0
             , ageclass <- if (age < 18) "child" else "adult"
             , gender <- switch( toupper(gender)
                               , "F" = "F"
                               , "V" = "F" # common mistake
                               , "M" = "M"
                               , "NB"
                               )
             )

m is now a set of rules that can be applied to a data.frame or tbl.

print(m)
#> Object of class modifier with 6 elements:
#> M1: 
#>   if (age < 16) income = 0
#> 
#> M2: 
#>   if (year < 25) year = year + 2000
#> 
#> M3: 
#>   if (cigarettes > 0) smokes = "yes"
#> 
#> M4: 
#>   if (smokes == "no") cigarettes = 0
#> 
#> M5: 
#>   ageclass <- if (age < 18) "child" else "adult"
#> 
#> M6: 
#>   gender <- switch(toupper(gender), F = "F", V = "F", M = "M", "NB")
# modify a copy of the table
modify(person_tbl, m, copy = TRUE)
#> # Source:   table<dcmodifydb_2573145> [4 x 7]
#> # Database: sqlite 3.38.5 [:memory:]
#>   income   age gender  year smokes cigarettes ageclass
#>    <int> <int> <chr>  <int> <chr>       <int> <chr>   
#> 1      0    12 M       2020 yes            10 child   
#> 2      0    14 F       2019 yes             4 child   
#> 3   2010    25 F       2019 no              0 adult   
#> 4   1010    65 M       2020 yes            NA adult

Note that the rules are executed sequentially, in the order that they are gven. For example the order of rule M3 and M4 matters: Rule M3 will change record 1 to a smoker, while rule M4 would set the number of cigarettes to 0. This is intentional: correction rules often have an order in which they have to be applied.

A nice properties of modifier rules, is that they can store extra metadata. They have a name, label and description that can be used to describe the intention and the why of a rule. An easy way of describing these properties is by exporting the ruleset to yaml and specify the rules using the yaml file.

export_yaml(m, "corrections.yml")

In the export yml file we can label and describe the rules, but also add new rules. Note that label and description are optional, but very much encouraged.

corrections.yml


rules:
- expr: if (age < 16) income = 0
  name: M1
  label: 'nochildlabor'
  description: 'Children are not allowed to work, so can not have income.'
- expr: if (year < 25) year = year + 2000
  name: M2
  label: 'longyear'
  description: 'Convert 2 digits year into 4 digits.'
- expr: if (cigarettes > 0) smokes = "yes"
  name: M3
  label: 'smoker'
  description: 'If you smoke cigarettes you are a smoker...'
- expr: if (smokes == "no") cigarettes = 0
  name: M4
  label: 'nosmoke'
  description: 'If you dont smoke, the (unknown) number of cigarettes is zero'
- expr: ageclass <- if (age < 18) "child" else "adult"
  name: M5
  label: 'ageclass'
  description: 'Derive ageclass using the age variable'
- expr: |
    gender <- switch( toupper(gender)
                    , F = "F"
                    , V = "F"
                    , M = "M"
                    , "NB"
                    )
  name: M6
  label: 'gender'
  description: 'Map the labels for gender to M/F/NB'

We can load these rules with:

m <- modifier(.file = "corrections.yml")
modify(person_tbl, m, copy = TRUE)
#> # Source:   table<dcmodifydb_3802595> [4 x 7]
#> # Database: sqlite 3.38.5 [:memory:]
#>   income   age gender  year smokes cigarettes ageclass
#>    <int> <int> <chr>  <int> <chr>       <int> <chr>   
#> 1      0    12 M       2020 yes            10 child   
#> 2      0    14 F       2019 yes             4 child   
#> 3   2010    25 F       2019 no              0 adult   
#> 4   1010    65 M       2020 yes            NA adult

modify translates the modification rules into SQL code and executes the sql queries on the database. For documentation or implementation purpose it can be useful to see the generated sql code, with the documented rules.

dump_sql(m, person_tbl, file = "corrections.sql")

corrections.sql:

-- -------------------------------------
-- Generated with dcmodifydb, do not edit
-- dcmodify version: 0.1.9
-- dcmodifydb version: 0.3.1
-- dplyr version: 1.0.9
-- dbplyr version: 2.2.0
-- from: '/tmp/RtmpVrsVJf/Rinst2ca4b29d44356/dcmodifydb/db/corrections.yml'
-- date: 2022-06-17
-- -------------------------------------


ALTER TABLE `person`
ADD `ageclass` TEXT;

-- M1: nochildlabor
-- Children are not allowed to work, so can not have income.
-- R expression: if (age < 16) income = 0
UPDATE `person`
SET `income` = 0.0
WHERE `age` < 16.0;

-- M2: longyear
-- Convert 2 digits year into 4 digits.
-- R expression: if (year < 25) year = year + 2000
UPDATE `person`
SET `year` = `year` + 2000.0
WHERE `year` < 25.0;

-- M3: smoker
-- If you smoke cigarettes you are a smoker...
-- R expression: if (cigarettes > 0) smokes = "yes"
UPDATE `person`
SET `smokes` = 'yes'
WHERE `cigarettes` > 0.0;

-- M4: nosmoke
-- If you dont smoke, the (unknown) number of cigarettes is zero
-- R expression: if (smokes == "no") cigarettes = 0
UPDATE `person`
SET `cigarettes` = 0.0
WHERE `smokes` = 'no';

-- M5: ageclass
-- Derive ageclass using the age variable
-- R expression: ageclass <- if (age < 18) "child" else "adult"
UPDATE `person`
SET `ageclass` = 'child'
WHERE `age` < 18.0;

UPDATE `person`
SET `ageclass` = 'adult'
WHERE NOT(`age` < 18.0);

-- M6: gender
-- Map the labels for gender to M/F/NB
-- R expression: gender <- switch(toupper(gender), F = "F", V = "F", M = "M", "NB")
UPDATE `person`
SET `gender` = CASE UPPER(`gender`) WHEN ('F') THEN ('F') WHEN ('V') THEN ('F') WHEN ('M') THEN ('M') ELSE ('NB') END
;