The Proxy class was originally created to insulate a user from TF dataset formatting and model restarts. As the project has evolved, this "wrapper" abstraction feels increasingly redundant and confusing. A cleaner approach would remove the Proxy class completely from the API and instead clearly document appropriate usage of the core PsychologicalEmbedding model.
One side-effect of removing the Proxy class is that that the Proxy class provides the only convenient method for computing the similarity between arbitrary points. This functionality would need to be provided elsewhere, perhaps in a standalone method.
The
Proxy
class was originally created to insulate a user from TF dataset formatting and model restarts. As the project has evolved, this "wrapper" abstraction feels increasingly redundant and confusing. A cleaner approach would remove theProxy
class completely from the API and instead clearly document appropriate usage of the corePsychologicalEmbedding
model.One side-effect of removing the Proxy class is that that the Proxy class provides the only convenient method for computing the similarity between arbitrary points. This functionality would need to be provided elsewhere, perhaps in a standalone method.