RegressionModelBuilder.approximate_fit#

RegressionModelBuilder.approximate_fit(X, y=None, progressbar=None, random_seed=None, *, fit_kwargs=None, sample_kwargs=None)[source]#

Fit a model using Variational Inference and return InferenceData.

This performs variational inference via pymc.fit, then draws posterior samples from the fitted approximation via Approximation.sample, returning an arviz.InferenceData compatible with the rest of the API (same structure as .fit).

Parameters:

Xarray_like | array, shape (n_obs, n_features): The training input samples. If scikit-learn is available, array-like, otherwise array.
yarray_like | array, shape (n_obs,): The target values (real numbers). If scikit-learn is available, array-like, otherwise array.
progressbarbool, optional: Specifies whether the fitting/sample progress bar should be displayed. Defaults to True.
random_seedOptional[RandomState]: Provides stochastic procedures with initial random seed for reproducibility.
fit_kwargsdict, optional: Extra keyword arguments forwarded to pymc.fit (e.g., {“n”: 10_000, “method”: “advi”}).
sample_kwargsdict, optional: Extra keyword arguments forwarded to Approximation.sample (e.g., {“draws”: 1_000}).

Returns:

az.InferenceData: Inference data of the variationally fitted model.