peal.generators.interfaces¶
Abstract interfaces every PEAL generator implements.
A generator is the model that produces counterfactual images for an
explainer or an adaptor. This module fixes the three capability levels the
rest of PEAL programs against: Generator (can be sampled from and
trained), InvertibleGenerator (also has a latent space one can
encode into and decode from, which is what the diffusion
autoencoders provide) and EditCapableGenerator (can run the whole
counterfactual search itself). GeneratorConfig is the pydantic base
that every concrete generator config extends.
Classes
A generator that can search for counterfactuals on its own. |
|
Base class of every PEAL generator. |
|
This class defines the config of a generator. |
|
A generator with an explicit, invertible latent space. |