Refactor tests to share data generation code

[markgoodhead]

Small PR to address #530

[codecov-commenter]

Codecov Report

Merging #531 (8d4a7dd) into main (91903c8) will decrease coverage by 0.05%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #531      +/-   ##
==========================================
- Coverage   86.84%   86.78%   -0.06%     
==========================================
  Files          32       32              
  Lines        2622     2611      -11     
==========================================
- Hits         2277     2266      -11     
  Misses        345      345              

Impacted Files	Coverage Δ
bambi/tests/test_built_models.py	99.00% <100.00%> (-0.04%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 91903c8...8d4a7dd. Read the comment docs.

[aloctavodia]

LGTM!

[tomicapretto]

Thanks for all the amazing work @markgoodhead !! One addition, I think that @canyon289 was actually saying we could have something like

bambi/bambi/tests/test_model_construction.py

Lines 289 to 304 in 91903c8

	@pytest.mark.parametrize(
	"family",
	[
	"gaussian",
	"negativebinomial",
	"bernoulli",
	"poisson",
	"gamma",
	"vonmises",
	"wald",
	],
	)
	def test_automatic_priors(family):
	"""Test that automatic priors work correctly"""
	obs = pd.DataFrame([0], columns=["x"])
	Model("x ~ 0", obs, family=family)

correct me if i'm wrong

Something like the following...

@pytest.mark.parametrize("method", ["nuts_numpyro", "nuts_blackjax"]) 
def test_logistic_regression_alternative_samplers(logistic_regression_data, method):
    model = Model("y ~ x", logistic_regression_data, family="bernoulli")
    model.fit(method=method, chain_method="vectorized")

On top of that, I would decrease tune and draws to something like 50 so it doesn't take too much time.

[canyon289]

Thanks for all the amazing work @markgoodhead !! One addition, I think that @canyon289 was actually saying we could have something like

bambi/bambi/tests/test_model_construction.py

Lines 289 to 304 in 91903c8

	@pytest.mark.parametrize(
	"family",
	[
	"gaussian",
	"negativebinomial",
	"bernoulli",
	"poisson",
	"gamma",
	"vonmises",
	"wald",
	],
	)
	def test_automatic_priors(family):
	"""Test that automatic priors work correctly"""
	obs = pd.DataFrame([0], columns=["x"])
	Model("x ~ 0", obs, family=family)

correct me if i'm wrong

Something like the following...

@pytest.mark.parametrize("method", ["nuts_numpyro", "nuts_blackjax"]) 
def test_logistic_regression_alternative_samplers(logistic_regression_data, method):
    model = Model("y ~ x", logistic_regression_data, family="bernoulli")
    model.fit(method=method, chain_method="vectorized")

On top of that, I would decrease tune and draws to something like 50 so it doesn't take too much time.

This is what I'm saying :)

[canyon289]

This is great too!

[canyon289]

These next couple could be parametrized so now theres only one test. We're really squeezing code lines out now but I hope its a good learning experience!

[markgoodhead]

Comments addressed 👍

[aloctavodia]

Thank you, @markgoodhead, this is great!