I second that question… I’m guessing that Sense’s data science environment and machine learning environment is custom built and proprietary, but built on top of a standard data science framework like R or Julia (lots of MIT grads), and a machine learning framework like Theano, Caffe or Keras. More glimpses and info on what they do internally in this excepted webinar.
For example, clustering by features at 1:20 into the video - you can see they use at least 17 features, likely more. Power and phase0 as visible axes for clustering in this screenshot.
And here are a couple of activities to consider doing while you wait:
-
Export your hourly data and compare against your power companies hourly measurements, assuming your utility enables that kind of access (mine does). That will give you a feel for Sense’s overall accuracy, plus hints at where it might be off from your revenue meter (hint - there can be places and times where your data might get lost communicating the Sense mothership)
-
Buy some TP-Link HS110s on Black Friday / Cyber Monday sales for use on outlet strips that power your router, servers (under 15A), switches, bridges, access point, AV equipment and other stuff that Sense won’t recognize for a long time to come. You’ll get quick identification and better accounting for some of your power usage right away.