workflows

High-level interfaces for amortized Bayesian workflows. BasicWorkflow is a good place to start; for ensemble-based inference see EnsembleWorkflow.

Examples

>>> import bayesflow as bf
>>> workflow = bf.BasicWorkflow(
...     simulator=bf.simulators.SIR(),
...     inference_network=bf.networks.FlowMatching(),
...     inference_variables=["parameters"],
...     inference_conditions=["observables"],
... )
>>> history = workflow.fit_online(epochs=20, batch_size=32, num_batches_per_epoch=200)
>>> diagnostics = workflow.plot_default_diagnostics(test_data=300)

Classes

BasicWorkflow([simulator, adapter, ...])	This class provides methods to set up, simulate, and fit and validate models using amortized Bayesian inference.
CompositionalWorkflow([simulator, adapter, ...])	This class extends the Basic Workflow to support compositional inference, allowing for the generation of samples conditioned on multiple datasets or compositional conditions.
EnsembleWorkflow([simulator, adapter, ...])	Ensemble variant of BasicWorkflow that trains multiple approximators jointly, allowing for flexible sharing of network components and training data.