I am trying to train a Keras model which has already been developed inside KNIME. Are there any main differences in using the DL Python Nodes for this as opposed to using the Keras nodes? Is is better to use the DL Python Nodes to move the Keras code into a KNIME workflow?
If you already have Keras code handy, it makes sense to use the DL Python Network Creator to implement that. The Keras nodes are there in case you want to define your network with a drag-and-drop, GUI config approach.
But either way, you should get very similar results. Take a look at this workflow, which uses both approaches to define the network.