As seen on Wikipedia, the simple double pendulum can exhibit chaotic motion, conditional on the proper initial conditions. There's a beautiful graph on Wikipedia that shows this phenomenon.
The question that comes to mind is:
Given our ability to easily simulate this kind of setup numerically, can we teach a machine learning system to predict
- whether the pendulum will flip (classification!),
- how long it will take a pendulum to flip (regression!)?
For now, running pendulum.py will generate a hdf5 file with time series data.