Workshop: Unit Testing in Python
- 1. Unit Testing Workshop It’s not that scary
- 2. 2 BEFORE THE WORKSHOP ● Download a code editor if you don’t have one ○ https://code.visualstudio.com/, or ○ https://www.jetbrains.com/pycharm/download/ ● git clone https://github.com/davified/unit-testing-workshop Windows users: ● docker run -it -v C:pathtounit-testing-workshop:/code -p 8888:8888 unit-testing-workshop ● Run all commands in the README up until: ○ nosetests --with-watch ● Configure your IDE (we’ll do this together)
- 4. ©ThoughtWorks 2019 Commercial in Confidence 4 Feedback Time Unit tests Jupyter notebooks IT’S ALL ABOUT FEEDBACK
- 5. 5 TEST PYRAMID Unit tests ML metrics tests Any other automated tests Manual tests Automate d Time required
- 6. How do we unit test?
- 7. 7 ANATOMY OF A UNIT TEST assert_that(do_something(input) => expected_output)
- 8. 8 ANATOMY OF A UNIT TEST assert_that(do_something(input) => expected_output) 1. Think about what your function is doing 2. Write the simplest possible test case to assert that
- 9. 9 ANATOMY OF A UNIT TEST def add(n1, n2): return n1 + n2 self.assertEqual(2, add(1,1))
- 10. 10 EXERCISE 1: UNIT TESTING A CALCULATOR 1. Code walkthrough: src/test_calculator.py and src/calculator.py
- 11. 11 EXERCISE 1: UNIT TESTING A CALCULATOR ● unittest module (built-in python module) ● How to write a unit test ○ File name must start with Test____ ○ Define a class that extends unittest.TestCase ○ Method name must start with test____(self) ● Assertions ○ self.assertEqual(expected, actual()) ● How to run tests ○ nosetests --with-watch
- 12. 12 EXERCISE 1: UNIT TESTING A CALCULATOR - YOUR TURN! ● Implement ○ subtract(a, b) ○ multiply(a, b) ○ divide(a, b) ● Challenge: write the tests first!
- 13. 13 GOOD TESTING PRACTICES ● Do the simplest thing possible ● Red green refactor ● Descriptive test names ● Pair program (+ ping pong)
- 14. 14 ANTI-PATTERNS / BAD TESTING HABITS ● Redundant tests ● Bad test names ● Empty tests (no assertions) ● Unspecific tests (e.g. Asserting that add(2, 3) isNotNone) ● Non-deterministic tests (causes flaky tests) ● Tautological tests (i.e. repeating implementation details in test cases)
- 15. 15 EXERCISE 2: UNIT TESTING + REFACTORING A JUPYTER NOTEBOOK notebooks/sklearn-nlp-pipeline-before-refactoring.ipynb
- 16. 16 EXERCISE 2: UNIT TESTING + REFACTORING A JUPYTER NOTEBOOK Instructions ● Read the notebook to get a sense of what it’s doing (it’s training a simple NLP model for sentiment classification) ● Sit back and relax
- 17. 17 ● Readable notebooks ● Automated tests -- no more visual eyeballing of long notebooks ● Smaller units → Separation of concerns ● Unit tests == living documentation ● Unit tests == safety harness ● Partition complexity (esp. if you TDD) ● Lesser bugs in production ● Possibility of refactoring → Ability to evolve to adapt to changing needs BENEFITS
- 18. 18 NEXT STEPS ● Learn by doing ○ http://codewars.com ○ Start with 1 unit test in your project codebase ○ Explore unittest docs: https://docs.python.org/3/library/unittest.html (they’re good) ● Resources ○ Unit Testing for Data Scientists - Hanna Torrence ○ Lots of tutorials on the internet
Editor's Notes
- #2: Add caveat on TDD
- #4: Enemy number 1 of software: complexity What adds to complexity? Long methods Cyclomatic complexity Etc. etc. Change We can’t refactor without unit tests Living documentation about our code Partition complexity Discover bugs early
- #5: Jupyter notebooks hit a roadblock at some point Unit tests can scale linearly
- #7: Let me share ThoughtWorks’ approach to machine learning
- #9: What to test: our functions What not to test: methods from a library
- #10: What to test: our functions What not to test: methods from a library
- #11: The goal of this exercise is to get you to familiarize yourself with the syntax; and see that it’s really not that scary
- #12: Demo for them. Delete add and go through this slide