Session 8: from the zoo to your own project#

Zoo model: any; the students choose

Goals#

  • Start a project from a zoo model and change it into the model of a question of your own.

  • Keep it reproducible and checked: a seed in the scenario, modelly lint clean, a test that runs the validation of the zoo model you started from, and one of your own.

  • Present a seeded experiment and one figure.

Before the session: Use Modelly in your project and Checking a model.

Exercises#

8.1 Copy a zoo model into a new folder and run it from there.

Solution 
modelly zoo list
modelly zoo copy vector_borne my_malaria
cd my_malaria
python -c "from model import VectorBorne, VectorScenario; m = VectorBorne(VectorScenario(rng=1)); m.run(); print(m.attack_rate())"

The commands are the same in bash and PowerShell. The copy is a single model.py plus a README; it imports only public modelly modules.

8.2 Copy the zoo’s validation test for your model from tests/zoo/ into your folder, change its import to from model import ... and run it with pytest.

Solution 
pytest test_vector_borne.py

It keeps passing as long as your changes leave the validated behaviour alone; when it fails, you know exactly which assumption you changed.

8.3 Make one change that answers your question (a new parameter in the scenario, a new rule in the step) and write a test that checks it on a case where you know the answer.

Solution

For example, bed nets that cut the biting rate a by a share coverage: add coverage: float = 0.0 to the scenario, multiply a by 1 - coverage where it is used, and test that r0() scales as (1 - coverage)^2, because the Ross-Macdonald R0 is proportional to a^2.

8.4 Run modelly lint on your folder and fix what it reports. Then profile one run.

Solution 
modelly lint .
modelly profile model.py:VectorBorne

MDL0xx findings are reproducibility bugs and must be fixed; MDL1xx are costs that grow too fast, worth fixing when the profile shows the step is slow.