====== Rabbit-In-a-Hat testing framework ======

**NOTICE FEB 2020:** the current documentation can be found on Github
http://ohdsi.github.io/WhiteRabbit/riah_test_framework.html


Rabbit-In-a-Hat can generate a framework for creating a set of [[https://en.wikipedia.org/wiki/Unit_testing|unit tests]]. The framework consists of a set of R functions tailored to the source and target schema in your ETL. These functions can then be used to define the unit tests. 

Unit testing assumes that you have your data in source format somewhere in a database. You should already have created an ETL process that will extract from the source database, transform it into CDM format, and load it into a CDM schema. The unit test framework can be used to make sure that your ETL process is doing what it is supposed to do. For this you will need to create a new, empty database with exactly the same structure as your source database, and a new empty database where a test CDM database will live. The framework can be used to insert test data into the empty source schema. You can then run your ETL process on the test data to populate the test CDM database. you can then use the framework to verify that the output of the ETL in the test CDM database is what you'd expect given the test source data.

===== Overview =====

These are the steps to perform unit testing:

  -  Create the testing framework for your source and target database schemas.
  -  Using the framework in R, define a set of unit tests
  -  Use the framework to generate testing data in the source data schema
  -  Run your ETL on the test data to produce data in the CDM schema
  -  Use the framework to evaluate whether the CDM data meets the defined expectations

It is advised to use [[https://www.rstudio.com/|R-Studio]] for defining your unit tests. One reason is that RStudio will automatically prompt you with possible function and argument names after you've only typed the first few characters.

===== Creating the testing framework =====

In Rabbit-in-a-Hat, have your ETL specifications open. The source data schema should be loaded from the White-Rabbit scan report, and the target data schema should be selected (usually the OMOP CDM v5). Go to //File → Generate ETL Test Framework//, and use a file name with the .R extension, for example ''%%MyTestFrameWork.R%%''.

===== Defining unit tests using the framework =====

Next, create an empty R script, and start by sourcing the R file that was just created:

<code>
source("MyTestFrameWork.R")
</code>

Be sure to run this command immediately to make the function definitions available to R-Studio.

==== Available functions ====

The test framework defines the following functions **for each table in the source schema**:

  * ''%%get_defaults_<table name>%%'' shows the default field values that will be used when creating a record in the table. At the start, these default values have been taken from the White-Rabbit scan report, using the most frequent value.

  * ''%%set_defaults_<table name>%%'' can be used to change the default values of one or more fields in the table. For example ''%%set_defaults_enrollment(enrollment_date = "2000-01-01")%%''.

  * ''%%add_<table name>%%'' can be used to specify that a record should be created in the table. The arguments can be used to specify field values. For fields where the user doesn't specify a value, the default value is used. For example ''%%add_enrollment(member_id = "M00000001")%%''.

The following functions are defined **for each table in the CDM schema**:

  * ''%%expect_<table name>%%'' can be used to state the expectation that at least one record with the defined properties should exist in the table. For example ''%%expect_person(person_id = 1, person_source_value = "M00000001")%%''.

  * ''%%expect_no_<table name>%%'' can be used to state the expectation that no record with the defined properties should exist in the table. For example ''%%expect_no_condition_occurrence(person_id = 1)%%''.

  * ''%%expect_count_<table name>%%'' can be used to state the expectation that a specific number of records with the defined properties should exist in the table. For example ''%%expect_count_condition_occurrence(person_id = 1, rowCount = 3)%%''.

One further function is available:

  * ''%%declareTest%%'' is used to group multiple statements under a single identifier. For example ''%%declareTest(id = 1, description = "Test person ID")%%''.

==== Defining unit tests ====

Using these functions, we can define tests. Here is an example unit test:

<code>
declareTest(101, "Person gender mappings")
add_enrollment(member_id = "M000000101", gender_of_member = "male")
add_enrollment(member_id = "M000000102", gender_of_member = "female")
expect_person(person_id = 101, gender_concept_id = 8507, gender_source_value = "male")
expect_person(person_id = 102, gender_concept_id = 8532, gender_source_value = "female")
</code>

In this example, we define a test for gender mappings. We specify that two records should be created in the ''%%enrollment%%'' table in the source schema, and we specify different values for the ''%%member_id%%'' field and ''%%gender_of_member%%'' field. Note that the ''%%enrollment%%'' table might have many other fields, for example defining the start and end of enrollment, but that we don't have to specify these in this example because these fields will take their default values, typically taken from the White-Rabbit scan report.

In this example we furthermore describe what we expect to see in the CDM data schema. In this case we formulate expectations for the ''%%person%%'' table.

We can add many such tests to our R script. For an example of a full set of test definitions, see the [[https://github.com/OHDSI/JCdmBuilder/blob/master/tests/HCUPETLToV5/HcupTests.R|HCUP ETL unit tests]]. 

==== Generate test data in the source data schema ====

After we have defined all our tests we need to run 
<code>
insertSql <- generateInsertSql(databaseSchema = "nativeTestSchema")
testSql <- generateTestSql(databaseSchema = "cdmTestSchema")
</code>
to generate the SQL for inserting the test data in the database (insertSql), and for running the tests on the ETL-ed data (testSql). The insertion SQL assumes that the data schema already exists in ''nativeTestSchema'', and will first remove any records that might be in the tables. We can execute the SQL in any SQL client, or we can use OHDSI's [[https://github.com/OHDSI/DatabaseConnector|DatabaseConnector package]]. For example:

<code>
library(DatabaseConnector)
connectionDetails <- createConnectionDetails(user = "joe",
                                             password = "secret",
                                             dbms = "sql server",
                                             server = "my_server.domain.org")
connection <- connect(connectionDetails)

executeSql(connection, paste(insertSql, collapse = "\n"))
</code>

==== Run your ETL on the test data ====

Now that the test source data is populated. You can run the ETL process you would like to test. The ETL should transform the data in ''nativeTestSchema'' to CDM data in ''cdmTestSchema''.

==== Test whether the CDM data meets expectations ====

The test SQL will create a table called ''%%test_results%%'' in ''cdmTestSchema'', and populate it with the results of the tests. (If the table already exists it will first be dropped). Again, we could use any SQL client to run this SQL, or we could use DatabaseConnector:

<code>
executeSql(connection, paste(testSql, collapse = "\n"))
</code>

Aftwerwards, we can query the results table to see the results for each test:

<code>
querySql(connection, "SELECT * FROM test_results")
</code>

Which could return this table:

^ID   ^DESCRIPTION TEST  ^STATUS  |
|101  |Person gender mappings  |PASS    |
|101  |Person gender mappings  |PASS    |

In this case we see there were two expect statements under test 101 (Person gender mappings), and both expectations were met so the test passed.