Double check "minimal attribute dict" fixture has the correct set of minimal arguments required
Set up the TestInitTestFit, TestTransform, and TestOtherBaseBehaviour classes to inherit the appropriate test classes and then copy the setup_class class method into each test class, changing transformer_name to the transformer you are testing (examples linked) .
Extra tests specific to this class should be added, e.g. In this case TestInit will need an additional check on the start_encoding arg. Where extra tests would apply to all child classes, they should instead be added to the parent test class. This class also contains a unique inverse_transform method that is not called by init/fit/transform, so needs to be tested separately - I would suggest moving this into BaseNominalTransformer.
Delete any unnecessary tests from original transformer test file (i.e. tests testing inherited behaviour that is now tested by the inherited test classes). An example would be. In general asserting calls is a red flag, as this tests the applied process rather than the desired outcome.
Debug the failing tests from the base test suite.
Bring the test file in line with best practices by insuring there are no remaining implementation tests
Bonus: If you identify any gaps in the testing, please set up new issues :)
Why?
Revamp tubular testing to make package more maintainable/modifiable
Depends on
What?
There are a few required changes for this, summarised below from (https://github.com/lvgig/tubular/issues/150)
Why? Revamp tubular testing to make package more maintainable/modifiable