BassModel.fit#

BassModel.fit(data=None, *, method='mcmc', progressbar=None, random_seed=None, sample_kwargs=None, **kwargs)[source]#

Fit the Bass diffusion model.

Thin wrapper around ModelFitter.fit(); see there for the full parameter reference.

Parameters:

dataxr.Dataset, pd.DataFrame, pd.Series, np.ndarray: Adoption counts over time. See to_bass_dataset() for formats.
methodstr: Method used to fit the model. One of "mcmc", "map", "demz", "advi" or "fullrank_advi".
progressbarbool, optional: Whether to show the progress bar. Defaults to True.
random_seedoptional: Random seed for reproducibility.
sample_kwargsdict, optional: Only used by the variational methods; forwarded to Approximation.sample.
**kwargs: Additional arguments forwarded to the underlying PyMC routine.

Returns:

xarray.DataTree: Posterior with parameters and deterministics (adopters, innovators, imitators, peak) plus a fit_data group.

Notes

After fitting, use standard ArviZ functions for posterior analysis:

import arviz as az

# Parameter summaries
az.summary(idata, var_names=["m", "p", "q"])

# Trace plots
azp.plot_trace(idata, var_names=["m", "p", "q"])

# Forest plots of peak adoption time
azp.plot_forest(idata.posterior["peak"], combined=True)

For posterior predictive sampling with new time points:

pp = model.sample_posterior_predictive(X=new_data)