Use bayesflow lre through hssm

[AlexanderFengler]

This is the first of two PRs and should serve as the basis for a decision on what kind of glue-code to own on this front.

In principle, we should be able to incorporate these types of networks natively as an option / addition into our overall pipeline.

Small bugfix along the way, bug was discovered while testing.

---

This pull request introduces integration with BayesFlow LRE and updates documentation and dependencies accordingly. The most important changes are grouped below by theme.

BayesFlow LRE Integration:

• Added an optional dependency group bayesflow in pyproject.toml to support BayesFlow integration, including required packages bayesflow and keras.
• Added a new tutorial on BayesFlow LRE integration: tutorials/bayesflow_lre_integration.ipynb is now included in the documentation navigation and build process. [1] [2]

Inference Method Handling:

• Simplified the inference method selection in the sample method of src/hssm/hssm.py by directly passing the sampler argument, allowing for more flexible integration of new inference engines like BayesFlow.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[Copilot]

Pull request overview

This PR lays groundwork for integrating BayesFlow likelihood-ratio estimators (LRE) into the HSSM workflow by adding an installable optional dependency set and exposing the new tutorial in the docs, alongside a small change to how HSSM.sample() forwards the chosen inference engine to Bambi.

Changes:

• Forward sampler directly as inference_method into bmb.Model.fit() from HSSM.sample().
• Add a bayesflow optional dependency extra (bayesflow, keras) to support the integration tutorial.
• Register the new BayesFlow LRE tutorial notebook in the MkDocs navigation and (non-)execution configuration.

Reviewed changes

Copilot reviewed 3 out of 5 changed files in this pull request and generated 1 comment.

File	Description
src/hssm/hssm.py	Adjusts how the selected sampler/inference engine is passed into Bambi’s fit() call.
pyproject.toml	Adds an optional dependency extra to enable pip install hssm[bayesflow].
mkdocs.yml	Adds the BayesFlow LRE tutorial to the docs nav and mkdocs-jupyter config.

Comments suppressed due to low confidence (1)

src/hssm/hssm.py:794

• hssm.HSSM.sample() now passes sampler straight through as inference_method to bmb.Model.fit(). There are existing tests that call sample(sampler="numpyro"), but they only assert that an InferenceData is returned, so they won’t catch regressions where the requested sampler is ignored or mapped incorrectly. Consider adding a regression check that the requested sampler/inference method is actually used (e.g., assert on metadata recorded in the returned InferenceData, or mock/spy on bmb.Model.fit to verify the argument value).

        self._inference_obj = self.model.fit(
            inference_method=sampler,
            init=init,
            include_response_params=include_response_params,
            omit_offsets=omit_offsets,
            **kwargs,

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[AlexanderFengler]

@digicosmos86 could you give a quick glance at the sampler routing logic and approve / request changes if you disagree with anything there?

Trying to get it into the release.

[digicosmos86]

LGTM! Seems like an easy integration with HSSM through a callable