Data Export

Digging Deeper: An Interview with kevin1

Our Community Manager recently talked with enterprising user kevin1 about his data-driven approach to using Sense.

kevin1If you’ve been to the online Sense community, one name has likely stuck out for its sheer ubiquity: kevin1. Kevin Kranen has been a Sense user since the early days and has become a great asset to the Sense community at large. Kevin is no fanboy singing our praises — though we’re certainly happy when he’s happy! An electrical engineer by trade, he has spent most of his career on chip and system design, and recently has taken an interest in data science and machine learning, making him an insightful and objective voice in the Sense community.

Kevin was also instrumental in the formation of a new Data Analysis group on the forum where users can chat about what they’re doing with their Sense data as well as homebrewed hacks for deeper integration with the vast Internet of Things.

I spoke to him recently about how he came to Sense and how he’s been using it lately. Things got pretty technical, so if you only know R as the 18th letter of the alphabet and Python as the nightmarish subject of your ophidiophobia (thanks, Wikipedia), don’t fret. Just join us over at the Community forum and ask questions

R: What drew you to Sense initially?

K: The ability to look at the economics of my electrical production and usage. But I was also intrigued by energy disaggregation via machine learning and how that would work. I had some experience in machine learning for image recognition, working with early versions of products that found faces and named them. Figuring out which devices were on or off via just the waveforms on the mains seemed like a more challenging and far reaching problem, but on the cusp of being solved. I had looked at a few products that all seemed to be trying to do the same thing, and chose the Sense product because the company seemed the most focused and best funded. One other draw for an electrical engineer — just the idea of having your house instrumented at such a fundamental and detailed level is really cool.

R: Can you tell me a bit about your background? You mentioned you’re an engineer.

K: As a high schooler, I was really big on building stuff, whether mechanical, architectural, with wood, or electrical. In school I was drawn to math and the sciences, especially physics which just seemed so intuitive to me. I thought about both architecture and engineering for college, and my guidance counselor steered me more in the direction of engineering.

In my time, in Cornell’s program, I squeezed in everything from Materials Science to Biophysics. But in the end I chose Electrical Engineering because that seemed to be the area at the time (early 80’s), where people were building the most interesting stuff. Computer science was a close second and I took a substantial number of Comp Sci classes in my 1-year follow-on MEng degree. After that I spent a lot of years designing and building chips, followed by validating and improving the really cool EDA (electronic design automation) software used to design chips.

R: How has this background informed your approach to using Sense?

K: I originally bought Sense because I was curious, as an electrical engineer, about what Sense could actually do and show, especially related to making decisions related to electricity usage. A couple years earlier, we added solar panels to our house and had purchased a couple electrical vehicles since then.

I really wanted to understand the economics of it all.

I really wanted to understand the economics of it all. When most people buy solar, including me, they base their decision on a very simple pricing and usage model based on a few big aggregate numbers. I really wanted to see if those numbers were still viable given all the utility pricing plan options and with our new usage additions, two Teslas.  

In an earlier incarnation, prior to discovering Sense and data science, I had tried to look at my energy numbers using downloaded data from my utility (with a device called Rainforest EAGLE that gives you a live feed from your electric meter), my solar utility downloadable data, and a spreadsheet. I quickly discovered a bunch of painful barriers: (1) None of my sources used the same time interval or units; (2) time zones and Daylight Savings are trickier than they look; and (3) my utility rates are incredibly difficult to compute for a variety of reasons.

Eventually I cobbled together a spreadsheet that could do very basic analysis with only one pricing scheme, but it was really a one-time only thing. Trying to do an update three months later using the same spreadsheet would still require 70% of the original effort I used to construct the first one. Not sustainable…

About a year later, I had completed my first data science class (about a month after I had Sense up and running) and decided it was worth trying to rebuild my calculator using R, with inputs from Sense, because Sense was already automating half of the problematic issues from my previous attempt. The only issue was how to get the data from the Sense app into R. Initially, and quite laboriously, I transcribed daily Sense data into a spreadsheet which I used to compare against my net meter data. I was getting ready to add my solar utility download into the calculator, but the addition of Sense’s web app changed my mind. Just about the time I had it working for an entire year, Sense began the beta of Data Export. My project was mothballed in favor of Sense’s export feature.

R: Can you tell me a bit about how you've put Data Export to use for your goals?

K: I’ve posted a few of my mileposts along my way to my energy cost calculator on the Sense Community forum, the scraper to extract data pre-export and my final calculator that includes both tiered pricing and ToU pricing. And I actually posted the costs associated with various alternatives based on my actual usage and solar production. I can actually say that I have saved about $3,200 per year from the combination of my solar install plus ToU pricing, plus charging EVs mostly at night. Not bad for a $15K investment.

I can actually say that I have saved about $3,200 per year....

 I also discovered real issues — our downstairs furnace was shutting down regularly because my cold-air return pipes weren’t large enough. I didn’t really notice until I looked closely at the Ecobee’s export trying to match it to Sense’s furnace fan info.

R: What programming languages or other tools are you using with Sense and why?

K: I use different tools for different needs with Sense. Mostly, I prefer R for analyzing already available data. Python for extracting data that I can’t access via Sense export or from the web app UI. And I have also used a Java/JSON interface that another forum member developed for accessing data from TP-Link smart plugs. Of course, just as I was fine tuning a smart plug monitor and exporter so I could pull power data for devices that had never been found by Sense, Sense started their smart plug beta — another project sidelined, but for all the best reasons.  

R: Overall, what do you see as the most useful aspects of Sense and what do you hope the future for Sense holds?

K: The beauty for new users is that they get a great turnkey capability that integrates power measurement and management from a number of different angles - whole house usage, solar production, detected devices, Hue lights, and smart plugs, all in one easy to use package. And even without full tiered or time of use pricing capabilities, a new user can get a solid picture of how much it costs to power different things in their house. I find the fundamental integration and simple to comprehend UI most useful, partially because I tried to glue together something similar with data from my utility supplier, solar supplier, etc. and it’s harder than it looks.

I’m looking forward to faster and faster learning of new devices as Sense analyzes more and more customer data.

R: Lastly, what draws you back to the Community every day what compels you to contribute so much?

K: A couple things: excitement about the product along with commensurate energy savings, plus idea generation.

I really do think we have to reduce our CO2 footprint globally, and the Sense Community is a hotbed of people mostly trying to do that. Sense is one of the core internet of things (IoT) tools for reducing energy usage at a personal level, along with smart thermostats, and renewable energy sources.  So I contribute, because I want to see everyone in the Community successful with their own personal campaigns.

From an idea generation perspective, I dabble and consult on data science in a number of areas. I like to turn big data into intuitive human insights.

I like to turn big data into intuitive human insights.

The Sense Community forum has been a real idea generator for me, mainly because it’s full of lots of people asking interesting questions. And often, an answer to one of their questions has much broader applicability.  For example, I developed some techniques to summarize heating, fan, and cooling cycles my Ecobees were calling for, to match against the results I was seeing in Sense, based on a question posed on the forum. That same chunk of code formed the basis for analyzing new voter entry and existing voter dropout across elections. But the solution to the problem was easier for me to visualize based on the forum question than while staring at voter rolls.


Kevin’s Projects:

Time of Use (TOU) Cost Calculator

Multiple Sense Monitor Data Aggregation


About the Author: Ryan LaLiberty is the Community Manager at Sense. You can find him on as RyanAtSense.