We should be able to TRAIN this!

@brian1, here’s what I suggest responded on the subject in another topic:

2 Likes

@becky, we do have something similar to what you mentioned planned :slight_smile: .

2 Likes

Just updating…Sense found (again) my old, removed garage door opener; not the new one I installed 2 months ago. I opened it via the LiftMaster app and noticed it wasn’t the correct device. I deleted it (again). This is the third time. When deleting a device you should delete the data it associated to be found too. Anyway…one more request for IoT learning/connectivity. This would be sooo easy to identify loads…

Sorry that your old garage door opener is being re-detected! This is an area of improvement on our radar.

In terms of additional IoT integrations and network detection-based detectors, they’re definitely something we have planned.

1 Like

Another example… My Sense has been running for about 8 weeks.

I charge my Tesla Model S just about every night at 12:00 a.m. I understand that Sense “knows” the Model S signature; but mine hasn’t been detected yet at my house, even though my Sense has been installed for about 8 weeks. I wish I could help it along my manually defining it.

1 Like

In 2 weeks I will have had my Sense for 1 year. 47.1% Unknown Devices, 38.1% Always On. 5.1% is for Refrigerators which cant be identified… so when you add it all up thats 90.3% of UNKNOWN stuff.

The remaining 9.7% is reasonably accurate but I dont spend the time anymore to correct the coffee pot from the waffle maker and such.

If my two forced air heaters would be detected I am sure the percentages would change substantially but this is the second heating season and they must cycle on and off 100 times a day but they have not been identified.

I have no lights, no garage doors ect.

I will also note that all of these comments for training have been mentioned by numerous other owners going back a year. Its very frustrating every new user has the same comments and we always get the same answer.

1 Like

I am a new sense user. Just installed it 2 days ago without issue. I am (and will continue to be) patient. To a point. In 2-4 weeks i would hope to see some accurate discoveries.

I agree with the rest. We should be able to train the sense based on our own usage. Walk around the house turn on a device, see it on the power meter, label it. I would be very surprised if SENSE can discover my Buderus modulating condensing boiler which provides my heat to 3 zones and domestic hot water all of which have cirulators. The boiler modulates its output based on outside temperature (outdoor reset) and even further modulates as it approaches the thermostat or DHW set point.

I understand that because this data is used by sense engineers to create “signatures” for the devices and for the betterment of the community, you wouldn’t want every user sending their specific unique data to you. But how about a choice? Implement the feature, and put a check box in the feature that says something to the effect of: “This is my unique device signature for a (fill in the blank/detailed description). I am notifying SENSE data engineers that it is unique to my installation and may or may not be of value to them.”

This would let your engineers know right off that there was human intervention in the acquisition of the device rather than being automatically discovered, and they can assign an appropriate “weight or value” to it and choose whether to use it or ignore it. This way the end user (me) has a far more accurate and intuitive device, in a much shorter time-frame.

Thank you.

1 Like

I am a new user, only had it for a few days but I am loving it. I knew going into this that sense was going to take a while to pick up stuff and I am somewhat OK with that.

My 2 cents for this is this: How about allowing us to highlight on the usage graph and giving that a Priority for Sense to find what it is.

Say I know when my AC kicks on and off, I would highlight that whole event then click that highlight to give it a Priority 1 so that Sense would know to “find” that quicker.

2 Likes

Here’s a little insight into the kind of analysis and math that makes Sense work. The real trick for Sense is being to operate without a “ground truth” measurement.

https://ac.els-cdn.com/S1877050917302417/1-s2.0-S1877050917302417-main.pdf?_tid=0e1abf26-1138-11e8-9764-00000aacb362&acdnat=1518579488_621831161cbabc3fe4469da114a9fdfd

2 Likes

Wow, I’ve had my Sense for a little over a week and this is very disheartening to read… :frowning:

1 Like

Sorry. 2 months later and nothing has changed. I could have said the same comments a year ago. Very little progress for my home.

Sense did just find “LIGHT 1” 2 days ago. I was optimistic that this would be the start of some new device identification. Sadly when I check the stats even though the device was just recently discovered it has been “off” for over 3 months with zero energy usage. This is very puzzling.

Also, I was really hoping that after 2 winters Sense would have identified either of my central heating units but it did not. They are very high (1000+ watt) devices that cycle on many times a day. It is now 75 degrees outside so it looks like I will need to wait until next year.

Sorry, but after over a year I just refer to mine as NONSense or NOSense. I has not done anything much and picks up little. Totally frustrating and nothing near what I anticipated or expected.

2 Likes

i’ve wondered if the devices that seem obvious to us that Sense isn’t picking up - could those have power fluctuations or other issues that keep the signal from being stable enough to be identified? or in the case of some people who have very little detection overall could there be a device on their system that’s so erratic that it confuses everything? i don’t have any machine language training or anything like that - just guessing from a layperson’s pov. my well pump seems clear as day to me from the first hour i plugged sense it, but while it’s identified minutia and components of many devices around the house, it hasn’t tagged the well pump, which I suspect could have issues. - also, it has id’d on hvac but not the other… a system i’ve had issues with since it was installed.

I would like to be able to “Train” Sense as well. But I am not confident it would be capable of reliably recognizing actual usage. My case in-point: My Sense learns/detects my Electric Vehicle, but then doesn’t reliably recognize actual usage/charging… despite a very distinct visual signature

UPDATE!!
Still the same, some things found early on are no longer detected.
Pool pump NEVER detected. Cannot detect the heat pumps like last year.
Sucks, its a bust!!

Update in March ‘18. Actually, no update. Same situation. Still 40% unknown. No resolution on my BMW i3 charging (can’t believe that the regular 10pm, 3 hour charging routine is impossible to recognize). Changed garage door opener 7 months ago and Sense still insists on finding the old one despite my deleting it at lease 3 times. Nothing new discovered. I have only one request Sense: put your resources in device detection. Everything else is a nice-to-have. That said I’m still glad I have it…but I suspect that there will be better solutions coming…

1 Like

It’s a Catch-22 for other solutions. One can’t improve machine learning without massive real datasets. And one can’t build massive real datasets without placing lots of probes in many different homes which implies other solutions would have to do better from day 1. Unless the other solutions uses some other core technology. Neurio and Smapee seem to have some data collection traction, but Neurio seems to be changing their business model to something different that involves less focus on detection. And Smapee doesn’t seem to be all-in on detection.

1 Like

Thanks for the input Kevin. I understand it’s hard to do data mining of big data without big data. That said, Sense seems to be more interested in driving their algorithm than actually detecting devices. Simplist example: I would happily give Sense access to my Nest account that drives thermostats. This should be a super easy way to monitor heater loads. Ditto for WEMO and lights/switches. Maybe they are working on this…maybe not. All that aside; detection is what I paid for and I’ve gotten 60% (or less) of that.

Then I think you would have been pleased by a couple of things they revealed in last Fri’s conf. call webinar. But I’ll leave denouement to the folks at Sense.

One question: What kind of heat do you have ? I have 2 forced air furnaces w AC, and Sense has found both furnace fans and compressors. But they are old school single stage non-varible units.

1 Like


I’m looking forward to hearing about that new info from Sense. I have two systems. One is a heat pump. Sense captures it occasionally but has missed the supplemental heat mode and the air circulation fan (see image 1500 Watts). On system 2, standard forced air gas furnace/AC, nada. Both are <5 years old. Someone told me here that variable speed motors (like my new garage door opener) are problematic for Sense. The garage door opener (LiftMaster) has integrated IoT and sends me an text every time it opens/closes. Sense could know exactly where/when loads are operating by linking to the various device mfg’s apps. Again, I like what they have done…IMO they need to do something different to keep making progress. I’ll search out that new info next. Thanks.

1 Like