Save intermediate state of models

[frneer]

Hacer cada cosa en pasos y guardar los modelos.
.save( https://www.tensorflow.org/tutorials/keras/save_and_load#new_high-level_keras_format

Original model

Pre-training

Alpha computation and set-up

Quantization

Cada step tiene

• batch size
• learning rate
• epochs
• Model
• Dataset

Metadata, hash de la configuracion que usaste.

Goal: no tener que volver a crear modelos intermedios, de manera de tener los parametros usados (pkl o json).

and the json metadata to have a history of that layer.

[frneer]

How to distribute keras models since they must be heavy? Let's think of alternitives