I recomment considering learning between test runs and I encourage you to train a relatively simple model for selection on top of http-archive and tagged data.

I recomment considering learning between test runs and I encourage you to train a relatively simple model for selection on top of http-archive and tagged data. "off the shelf" machine learning makes it pretty easy to create very robust selectors. I gave a talk about it in GDG Israel and was supposed to speak about it in HalfStack that got delayed cancelled because of COVID19 - but the principle is pretty simple. It's amazing how much RoI you can get from relatively simple models of reinforcement learning. Here are some ideas: https://docs.google.com/presentation/d/1OViIwDJJw1kjVJH5Z2N5chz5Usw3kOf41hCiwZXu0ng/edit?usp=sharing https://docs.google.com/presentation/d/1OViIwDJJw1kjVJH5Z2N5... Good luck :]