Manual training loop instead of


instead of using is there anyway to do custom training loops…

Just like this in tensorflow → Custom training: walkthrough  |  TensorFlow Core

I spoke to the NengoDL devs, and according to them, you should be able to use the reference to the Keras model to do what you are asking. For each NengoDL simulator object, the sim.keras_model contains a reference to the underlying Keras model that is used in the TensorFlow training, so in theory you should be able to do a custom training loop using that model. However, using the Keras model reference directly does mean you will lose all of the overhead that NengoDL does to integrate the Keras model with Nengo - and this is primarily the generation of input data, and the collection of output data from the Keras model. In essence, if you were to take this approach, you’d need to perform all of these things manually. You can get an idea of how this is done from within the NengoDL simulator code base (see these lines).