Machine Learning with LunchBoxML

Earlier this summer, I previewed some of our research on machine learning which focused on potential applications for the building industry. I am now pleased to announce that these explorations have been released as a new extension to our LunchBox tools for Grasshopper: LunchBoxML.

LunchBoxML exposes new open source components built on a popular machine learning library called Accord.NET. With LunchBoxML, a Grasshopper user can implement and extend popular regression algorithms, clustering models, and neural networks within the computational design workflow.

Download LunchBox for Grasshopper

Visit the source code at Bitbucket

LunchBoxML for Grashopper.

Why Machine Learning?

A great deal of mystique and hype surrounds the term ‘machine learning’. The term itself might evoke images of computers eating up information as they gain sentience leading to a Matrix-style future where leather-clad hackers must eventually defend humanity from the new machine overlords.

The reality of machine learning is much more practical…even a little mundane. However, this is not to say that machine learning does not offer some exciting possibilities.

A sample file showing a basic linear regression setup with LunchBoxML

Simply put, machine learning is about creating an algorithm that can learn from some quantified ‘experience’ so it can then make predictions based on this experience. Online shopping is a rather pervasive implementation of this concept: An algorithm will take information about past purchases in order to make recommendations to the shopper about what they might like to buy next. For the building industry, we can envision predictive models being used for product specification, identifying occupancy trends, and helping to optimize building performance.

Yet for the all the novel implementation we can do with a neural network or a Bayesian model, it is the data itself that ultimately provides the value for an organization. As more data is gathered over time, the better the predictions by the algorithm.

A sample file of nonlinear regression to describe a surface with random training points.

It is for this reason I often ask my clients in the building industry the question: “what data do you have that no one else has?” or “what data do you have more of than anyone else?” Consider the possibilities for a healthcare design firm that can use patient data to help a clinic determine future expansion potential. Or imagine a developer that is able to utilize demographic trends combined with data from previous projects to best situate their next commercial venture.

To put things another way: while we are excited to be releasing LunchBoxML, it will ultimately be your data and ingenuity that will deliver on the promise of machine learning.

Are you interested in using machine learning in your business?