[Tensorflow] Implementing Different Training Methodology

[AnuragKr]

Hello,

Objective -- I want to implement a new training methodology for EnergyFitting using an Online Sequential Extreme Learning Machine (OS-ELM) Algorithm rather than Feed Forward Neural Network which doesn't require iterative training.

The problem is that Descriptor and Energy Fitting currently have different computation graphs but the same training method for weight optimization using Adam Optimizer. And Descriptor can't be trained independently as it is an intermediate step that doesn't have any of its own data for optimization.

So How to implement different training methodologies under the same session of Tensorflow training for Descriptor and EnergyFitting?

It would be a great help if you can provide suggestions, How to go about it or if Is it a viable option to try?