We should be able to TRAIN this!

technicalsupport

#1

I would like to train this thing. Turn something on, mark it on the graph, and define it.
I have a pool pump that is always on, I could turn it off and on, high and low speeds and mark it.
What could be easier?? But it never gets it. Same with most appliances, lights, fans, etc. Instead we play “peeka boo” waiting for “IT” to say what a device is, maybe. You should add this as a function for those of us willing to go through it. I would be willing to go through this.


#2

We definitely want users to be involved in the device detection process. If we felt that manual training would a good use of your time, we would work on implementing it (see: http://blog.sense.com/articles/training-sense/ for more information).

That being said, we are looking for more ways to have users involved in the device detection process. For example, we added the ability to let us know when a ‘Device is not on’. If you see a device on when it shouldn’t be, you can mark it as such (in the device details screen under ‘Report a problem’). This information is relayed to our data science team so that they have more information to improve our detection algorithms. Renaming devices is another way to be involved in the device detection process.

We are working on more features along these lines so that users can help our data science team and our machine learning algorithms improve device detection.


#3

Tried to email you back but it was rejected, need to fix that. ANYWAY, as I am a retired Science teacher, and I bought this thing to monitor my electrical usage, (particularly the devices it cannot find) I am the best judge of my time usage. As I have A LOT of time to devote to finding the mysteries of device detection, I would LOVE to have that option. It could pay you back in better algorithms for this Sense device. ALSO, I have MANY UPS backups connected to sensitive electronics equipment, is Sense ever going to be able to find these? ANY of these?
Thanks.


#4

Jack,. I find this also a much needed feature, it would seem to me that having Sense online while walking around and turning devices on and off to detect them would be a major short cut in the Sense learning curve plus allow humans to ID said devices along with model numbers and so on that should be added to the list of known devices ( power signatures ) which in turn should add greatly to the over all speed of learning on Senses part. My old TED5000 had a very similar feature which allowed me to bring it up to speed pretty quickly.

BRW


#5

Thanks for the suggestion. The way our device detection processes work, we’ve found that turning devices off and on and labeling them in that manner to not be the most effective way to accurately identify devices. That being said, there are several ways that user input contributes to our algorithms (mentioned above) and we’re working on others!

With regards to UPS devices, the fact that they are connected to a UPS shouldn’t prevent Sense from detecting them.


#6

One sorely missing but simple feature: identify devices as two phase (220V). Last season Sense identified my in-floor radiant heat and tracked it perfectly. But over the summer it decided it was the toaster oven, then lost it. Since the heat is 220V, these should be easily distinguishable. But sadly, now that the heating season is here again in Ohio, it’s MIA. Also MIA are two exhaust fans for twin hot water heaters, which were among the first things Sense identified last January. But since there were two of them, I think Sense gave up in its confusion. I should be able to say “I have two identical units, don’t be confused” (goes for garage door openers too). So there are plenty of ways short of full training that we could be involved, but aren’t yet. Just my perspective.


#7

You keep saying that it is not the most efficient way to identify. But TRAINING/waiting/hoping really is not working. We are waiting/waiting/waiting and hoping/hoping/hoping for months and it is a very disappointing outcome.


#8

#9

If a user really, really wants to work on training Sense. But you better understand the works by Kelly:


and Mauch:

before you apply.


#10

Or we could take the “common sense” approach. Turn everything OFF at the fuse box. And then turn on an appliance, on a branch, one at a time and mark it as a device. What could be easier? Then wait for other “discoveries” as they occur. This would help those of us, who are most interested in getting this DONE. And PERHAPS help others who are waiting for “discoveries” by providing fresh data. After waiting MANY months, I am still waiting for Sense to “Discover” my pool pump which is a big power consumer for me.


#11

Sense really needs to make a FAQ page or something that explains how machine learning works, in particular why you can train with training sets but not necessarily the way people are expecting. These posts seem to come up quite often and it’s a mixture of frustration and how it SEEMS like it should work, despite the fact is doesn’t really work like that.

I mean I get it, you see the same thing cycle all the time on and off with the same wattage so the reaction is “It’s right there!!! I can see the pattern!!” you don’t think “I wonder which existing neural net pattern this will match”


#12

I’ve done a few small repetition oriented things that seem to have helped:

  • Turned my furnace blowers on for 5 minutes every hour - definitely helps with identification plus seems to have lightened our heating and cooling bills.
  • Put 5 clusters of “Always On devices” on power meter / timers so my late night “Always On” drops lower than it did originally, plus I can see the power flowing to these clusters when they are on.

#13

Sometimes I wonder what the effective difference is between Sense seeing something and the user seeing something. I can identify every item in my house, whether Sense can or not. I understand everyone’s home isn’t as simple as mine, but if I can see a pool pump then what does it matter whether Sense has parsed it out or not?

Want to do something meaningful with that data? Ask Sense to work on allowing us to download our data in X minute increments.

Want to make changes to how that particular appliance operates? Make it smart, by adding a timer or using a WiFi-based control like Ecobee.


#14

From an electrical engineer’s perspective, I realize that there are huge differences in how a refrigerator powers up when it kicks onto cooling by itself and shuts down by itself compared to applying power and removing power from the refrigerator. Basically, the power usage of the refrigerator, graphed, looks VERY different when you kill the power and apply power than it does when the refrigerator turns itself on and off by itself. I understand that these types of devices can’t be force fed to Sense… but I think people would be much happier if we can help carefully inject data to improve the device detection time.

The heater fan in my bathrooms… those would be perfect candidates for “force feeding” information to sense because their normal operation has to do with turning the switch on and off… but there would have to be a way to trigger a learning mode… a set amount of time the heater would be left on for (30 seconds… a minute… what ever Sense Team determines the minimum to be) and then turning it off. For devices like a bathroom heater fan, or a pool pump… things that are meant to be power cycled, I have to agree with people asking for ways to help Sense learn these devices.

There are times when I know exactly what device turned on even looking at the general graph and if I had a way to select the start point and send some data to the Sense database, these things could be used to improve device detection.

I have to agree with the 240V vs 120V detection. Sense should already know that my refrigerator is a 120V appliance because it should have detected that energy usage on one of the two 120V legs and not the other. I would have a much easier time identifying devices if I could tap on the detected device and at least find out if it is registering as a 240V or 120V source.


#15

Amen to this. I know what my stuff is and want to train the models myself.


#16

I’ll just add my 2¢. After 1 year I feel like I’ve gotten 60% of an energy monitor (40% is unknown). Clear, big, predictable loads (‘17 BMW i3, portable Hot Tub heater, etc) have not been found (nor have many smaller known-to-me devices) despite numerous interactions with the “technical” folks. Promises. I’ve made several suggestions etc with “thanks…working on that” responses. Many, many posts focus on that singular issue…poor detection. We all fit into 1 of 2 categories: 1) happy with % detected and what richer features or 2) better detection regardless of how we get it and specifically not willing to wait another year for more of the same”. I won’t send it back and ask for a refund because 60% is ok. Just don’t expect me to be patient any longer or to offer good reviews when I’m asked.


#17

Well said. I’m in the same camp!


#18

Thanks for sharing @mstraka606. I’m sorry to hear that you have 40% unknown. Unfortunately it takes some time to collect enough data and develop algorithms for some devices. The team is working hard to expand detection coverage across more and more devices. That’s not to say that you should review a certain way, but just to say that we hear you, understand the frustration, and are working to get everyone’s % unknown down.


#19

Ben …you know what would help speed this process ?..allow your user base to train Sense via known devices.


#20

I don’t know if this has been suggested. I understand what’s being said about identification, , there is also the thought that users could mis-identify a device and skew results rather than helping.

What if users could access a list of devices that sense has a good handle on and click off a list of devices that they have. It could give sense a lean on that direction and ID known devices more quickly.

My main frustration, and from my reading many people feel the same, is the waiting . LOL.

For example, I have two heat pumps, upstairs and downstairs. it ID’d one weeks ago, but not the other and I wonder why. They run basically at the same time and it seems like the signatures would be similar. If I could tell sense, yes, I have two similar devices, would it go ahead and pick up the second one.?