models.BayesStacking
Bayesian stacking: a posterior for the stacking weights, with a
Usage
models.BayesStacking()Dirichlet prior, sampled by NUTS from pointwise held-out log densities (Yao et al., 2018). stacking_weights gives the optimum alone.
Parameters
concentration: list of float-
Dirichlet concentration, one per model (default 1, uniform).
chains: int = 4, 1000, 1000tune: int = 4, 1000, 1000draws: int = 4, 1000, 1000seed: int = 0
Methods
| Name | Description |
|---|---|
| fit() | Samples the weights. |
fit()
Samples the weights.
Usage
fit(lpd)Parameters
lpd: list of list of float- One list per model, one held-out log density per observation.
Returns
StackingFit