BVARResult._from_fit#

classmethod BVARResult._from_fit(fit, model)[source]#

Assemble the public result from a raw fit and the model that produced it.

The single construction point for BVARResult. BVAR.fit() runs the conjugate update through _BayesianVectorAutoRegressionModel._fit_conjugate(), which returns a _VectorConjugateFit holding only what the numerics produced, and this method joins that with the specification the model holds – the panel, the labels, the order, the trend, and the prior’s label – to make the frozen public record. Keeping assembly here rather than in __init__ keeps the result a plain dataclass with no logic of its own, and keeps the raw fit free of anything the sampler does not need.

The mapping is one-to-one, with a single rename: fit.coefficient_stack becomes coefficients. Arrays are passed through by reference, not copied; both containers are frozen, so the aliasing is safe.

Nothing is validated. The fit is trusted to be consistent with the model that produced it – same k, same p, same trend, draws already laid out in the model’s design order – because _fit_conjugate is the only producer and it guarantees exactly that. Passing a fit from a different model or specification produces a result whose arrays disagree with its labels, silently.

Parameters:

  • fit (_VectorConjugateFit) – The raw conjugate posterior output: posterior-mean blocks, the (S, w, k) and (S, k, k) draws, the closed-form log marginal likelihood, and the sample accounting.

  • model (_BayesianVectorAutoRegressionModel[BVARResult]) – The specification the fit was computed under. Supplies endog, names, order, trend, and, through model.prior._label(), prior_label.

Return type:

BVARResult

Returns:

The assembled BVARResult, sharing the fit’s arrays.

Example

Round-tripping a fit through the assembler is a pure field mapping:

>>> import numpy as np
>>> from cultivars.multivariate.large_dim.bayesian import BVAR
>>> from cultivars.multivariate.large_dim.bayesian import BVARResult
>>> model = BVAR(np.random.default_rng(0).standard_normal((60, 2)), order=1)
>>> fit = model._fit_conjugate(n_draws=20, seed=0)
>>> res = BVARResult._from_fit(fit, model)
>>> res.nobs == fit.nobs and res.n_kept == fit.beta_draws.shape[0]
True
>>> res.beta_draws is fit.beta_draws and res.endog is model.endog
True
>>> res.prior_label
'niw(l1=0.2, l3=1, l4=100)'
On this page