Add build check function, on prior predictive and plot prior

[tjburch]

Added a check that the model has been built in order to call prior_predictive. Since the same check is being called for this and plot_priors, I moved it out to a function that raises a ValueError if unbuilt.

Also added a single test that creates an unbuilt model and looks for that ValueError.

Closes #595.

[codecov-commenter]

Codecov Report

Merging #605 (7824d72) into main (efac57d) will increase coverage by 0.03%.
The diff coverage is 100.00%.

❗ Current head 7824d72 differs from pull request most recent head f9f96f2. Consider uploading reports for the commit f9f96f2 to get more accurate results

@@            Coverage Diff             @@
##             main     #605      +/-   ##
==========================================
+ Coverage   82.81%   82.85%   +0.03%     
==========================================
  Files          38       38              
  Lines        3020     3027       +7     
==========================================
+ Hits         2501     2508       +7     
  Misses        519      519              

Impacted Files	Coverage Δ
bambi/models.py	85.71% <100.00%> (+0.12%)	⬆️
bambi/tests/test_model_construction.py	97.29% <100.00%> (+0.05%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[tomicapretto]

bambi/bambi/models.py

Lines 883 to 887 in 7824d72

	if self.backend is None:
	raise ValueError(
	"The model is empty. "
	"Are you forgetting to first call .build() or .fit() on the Bambi model?"
	)

The above is not exactly the same type of check but it's very similar. There will be cases where changing the priors won't change the graph, if you still use the same family of distributions for example, but there will be cases where it would change the graph. So I think we should consider the check here as well. What do you think?

[tjburch]

Good catch.

That checks on a backend, which is set within the build function only, so it looks to me like in all cases where the model is built the backend should exist, so doing the check on being built be identical to checking if the backend exists. So I'll just run the same check. If I'm not thinking of some edge case let me know.

[tomicapretto]

Does it look like the failing test is due to numerical issues?

[tjburch]

Yeah, looks like some spurious divergences. Thanks for triggering it again.

[tomicapretto]

Thanks @tjburch :D