The last framework we are going to look at is TF-Agents, a relatively new but up-and-coming tool, again, from Google. It seems Google's approach to building RL frameworks is a bit like RL itself. They are trying multiple trial and error attempts/actions to get the best reward—not entirely a bad idea for Google, and considering the resources they are throwing at RL, it may not unexpected to see more RL libraries come out.
TF-Agents, while newer, is typically seen as more robust and mature. It is a framework designed for notebooks and that makes it perfect for trying out various configurations, hyperparameters, or environments. The framework is developed on TensorFlow 2.0 and works beautifully on Google Colab. It will likely become the de-facto platform to teach basic RL concepts and demo RL in the future.
There are plenty of notebook examples to show...
