Feature: Implement Encode/Decode Methods

[DiogenesAnalytics]

Problem

Currently there are no "sub-models" being created for the encoder and decoder. This is due to the complexity involved in determining which layers to use as input/output for each model. Regardless of the eventual implementation strategy chosen, the BaseAutoencoder class should have both encode() and decode() methods.

References

• https://gist.github.com/DiogenesAnalytics/399a96031ac360da5c34a000f3eae4ea#file-minimal_autoencoder-ipynb

[DiogenesAnalytics]

Class: MetaLayer

Might make sense to refactor code to use a MetaLayer class to store layer type/params:

from dataclasses import dataclass
from collections.abc import Sequence
from typing import Any
from typing import Dict

from keras.layers import Layer
from typing_extensions import override

@dataclass
class MetaLayer(Sequence):

    layer: Layer
    params: Dict[str, Any]

class Encode(MetaLayer):
    pass

class Decode(MetaLayer):
    pass

References

• https://docs.python.org/3/library/collections.abc.html