SINDyPy workshop

While the tension between theoretical modeling and fully data-driven applications is still ongoing (especially in our institution, thanks to the recent debate held during a Sociomeeting), the most pragmatic approach is simply to ask: how can we combine both perspectives to better adapt to today’s challenges? One nice middle ground is to extract theoretical equations from data observations— and figuring out the best way to do this has become quite a productive area in recent years. One popular method is SINDy (Sparse Identification of Nonlinear Dynamics), which aims to approximate a series of observations with a (possibly nonlinear) differential equation, under the constraint that the resulting expression is sparse within the space of all possible term combinations.

There’s also a Python package that lets you apply this method to your own data. I’ll go over some practical examples and give a quick tour of what the package can do (most of the features I have not had the opportunity to apply, so we’ll be discovering them together). Since my main area is Graph Theory, I’ll also talk briefly about an extension of this method to graphs, called SINDyG. I’ll focus on a few problems I encountered while using it, and share a (not very efficient) solution I came up with to work around them.