models

Description

Models and handles to define and store concept embeddings. This encompasses

• model architecture definitions

• an implementation independent format for representing a concept embedding obtained from a linear model, i.e. with a concept vector (the weights) and an offset (the bias)

• wrappers to extend and (re-)attach to DNNs

Sub-modules

concept_models	Concept model architectures and train eval handles for concept embedding analysis.
embeddings	Unified representation of concept embeddings with standard processing.
model_extension	Wrapper classes to slice and extend torch.nn.modules.