Comment by huula

Cool work! I'm very interested in this topic. Just wondering, how good does it generalize your training data other than just remembering strict input-output mapping?