Somehow in the last few months I’d missed the serious update of Michael Schmidt’s Eureqa package for symbolic regression. Now available for Windows, Linux and Mac platforms, and featuring a very nice cloud integration for additional processing, this looks like a ground-breaker for usability and understandability in Symbolic Regression applications.
Here’s an introduction video. Michael does a much better job explaining it succinctly than I’ve seen before:
Interestingly, I stumbled across Michael’s update because I was writing to Cosma Shalizi, fishing for feedback from the Very Important Thinkers attending the Ockham’s Razor Workshop regarding (modern, multi-objective) symbolic regression of the style Michael’s project embodies. A lot of these same notions of Pareto Symbolic Regression were developed originally by Mark Kotanchek, Katya Vladislavleva and Guido Smits from Dow Chemical and Evolved Analytics, recently released in DataModeler for Mathematica.
It’s fascinating to me how little professional attention crosses in either direction between the machine learning/statistical theory folks and genetic programming folks. Important work in both fields is essentially invisible across that divide. As the years go by the dichotomy is getting somehow deeper, to the point where I expect they’re just going to run into one another headed the other way, when they both circumnavigate the network of All Possible Approaches to Science and Engineering in their rush apart….
Note, as clarification: I am not including any chapters on symbolic regression in the book. Symbolic regression is an amazing and rapidly maturing field, and I count it pretty much “done” with the release of DataModeler and Formulize. From here on, it’s a field in its own right, not least because the tricks and techniques useful in addressing quantitative modeling projects like these are quite different from the qualitative and structural modeling projects we’re doing in the book.
Still, download and play with Eurequa II, and see what you can understand.