I've got some classical training data, and, for the moment, each of these will have its own embedding parameters defining how it is inputted into a certain ansatz. What I'd like is a circuit in two parts:
encoding circuit --> classifier (--> readout)
where the classifier is trained, but is the same for every input where the encoding circuit uses the parameters for each input -- i.e. I want to train the encoding as well as classification.
I can't work out the logistics of how to do this in tfq. I understand the MNIST examples, where each input has a predefined mapping to a quantum state, and then a PQC is trained. There's also the reinforcement learning example where you have a different set of parameters reuploaded between alternating PQC layers, but again this isn't quite the same.
tfq.resolve_parameters looks like it might help me but I'm not really sure how to integrate it into the model code
Any insight appreciated :)
