This tutorial teaches you to build your own physics-inspired reinforcement learning (RL) environment featuring simulations of a few ions manipulated by laser pulses.
We consider a standardized setting which can be summarized with some pseudo code integrated in the standard picture for the RL paradigm:
- Build the RL environment interactively in the IPython notebook Ion Trap Environment.
- Check out how a simple tabular projective simulation agent works under the hood by looking at the fully commented, short code in ps.py
- Try it out yourself and let an agent learn in your newly built environment by following the instructions in the IPython notebook Reinforcement Learning for Ion Traps.
- Explore what else you can do with this environment when using different agents, maybe even deep RL.
- Find more ingenious ways of translating physics problems into RL environments.
Before we get to the meat, here is some preliminary information.
numpy and scipy are the most important python extensions which we are going
to rely on in this tutorial.
Some jupyter notebook extensions are useful. So if you don't have any
extensions, install them as described
here
and here:
pip install --user jupyter_contrib_nbextensions
pip install --user jupyter_nbextensions_configurator
jupyter nbextensions_configurator enable --userThis allows you to activate certain packages that may come in handy.
Specifically for the exercise Ion Trap Environment Collapsible headings is
a plus as well as Freeze.
Strictly speaking however you don't need any. They are just for the experience.