Training a chatbot; we thought of a wrinkle that would let us use DRL to improve performance over pure data trained ones. It was hard to get it going, but once

Training a chatbot; we thought of a wrinkle that would let us use DRL to improve performance over pure data trained ones. It was hard to get it going, but once we figured out the code it did add vs pure supervised.