Serialization¶

Model persistence: MMMSerializer save/load for core and extended models.

Model serialization utilities for BayesianMMM.

This module provides functions to save and load BayesianMMM models, separating the serialization logic from the main model class.

class mmm_framework.serialization.MMMSerializer[source]¶

Handles save/load operations for BayesianMMM models.

This class encapsulates all serialization logic, keeping the main BayesianMMM class focused on modeling.

Examples¶

>>> # Save a model
>>> MMMSerializer.save(model, "models/my_mmm")

>>> # Load a model
>>> model = MMMSerializer.load("models/my_mmm", panel)

classmethod save(model, path, save_trace=True, compress=True)[source]¶

Save a BayesianMMM model to disk.

modelBayesianMMM: The model instance to save.
pathstr or Path: Directory path where the model will be saved.
save_tracebool, default True: Whether to save the fitted trace. Set to False for a smaller save file if you only need configurations.
compressbool, default True: Whether to compress the trace file with gzip.

classmethod load(path, panel=None, rebuild_model=True)[source]¶

Load a saved model from disk.

pathstr or Path: Directory path where the model was saved.
panelPanelDataset | None: Panel data to use with the loaded model. Must be compatible with the original data (same channels, controls, dimensions). REQUIRED for core (BayesianMMM-family) saves; ignored for extended-flavor saves (BaseExtendedMMM models carry their arrays in the pickle and need no panel).
rebuild_modelbool, default True: Whether to rebuild the PyMC model. Set to False if you only need access to the trace and don’t need to make predictions.

classmethod save_trace_only(trace, path)[source]¶

Save only a trace to a file.

classmethod load_trace_only(path)[source]¶

Load a trace from a file.