Hi. I would like to perform offline, imitation adversarial learning(GAIL) in my project. It is my understanding that I need to provide data from the teacher and then use it in training, providing analogous data from the student for discriminator to compare. I have motion capture data (tracking point positions) and I can provide analogous tracking data live from the agent in any frame directly to mlagents. I want to perform GAIL training on this data i.e. I want the discriminator to compare the selected tracking points from the data and the agent's body. How can I do this in ML-Agents?