allennlp.models.ensemble#

Ensemble#

Ensemble(self, submodels:List[allennlp.models.model.Model]) -> None

An ensemble runs multiple instances of a model and selects an answer from the subresults via some ensembling strategy.

Ensembles differ from most models in that they do not have a vocabulary or weights of their own (instead they rely on the vocabulary and weights from submodels). Instead, the submodels are trained independently and the ensemble is created from the result.