Another approach to Manually Training Sense


Ok, so a lot of people want to be able to train sense (myself included). I understand the reasons why it is not as ideal as we’d like it to be. So how about instead of adding a feature to the app to tell sense what is running. What if we could (at the least) tell sense what is NOT running.

and by this I mean, to help with Sense’s confusion between devices. Sense frequently mixes up my dehumidifier and the blower motor for my HVAC. the only way for it to distinguish between them is if I start the dehumidifier first, and then start the blower motor. doing it the other way around, will put the blower motor under dehumidifier, and the dehumidifier under “other”.
edit: and sometimes, like just now after a minute of running, Sense decided the blower was no longer running and filed the continuing power usage under other.

While selecting a point in time when I started the device and telling Sense that a device is running from that time may not work so well. Surely being able to tell Sense that a device it thinks is on, is not.

Like in this case, the blower motor and dehumidifier seem to have the same fingerprint. when one starts, Sense assumes it is the dehumidifier first. then when the other starts, Sense, unable to choose the dehumidifier twice, chooses the blower motor. I think it would be useful to be able to start the blower, and then when Sense says the dehumidifier is running, I could tell it, no the dehumidifier is not running. try again.

and perhaps telling Sense that a certain device(that is already detected) is running. This one would be a little dangerous though since people might apply it to components of a machine by accident/ignorance and mess up the data.

I mean I don’t know how all this works, so maybe this idea is not a good one. but maybe it has some merit? what do you think?


Hi @Grandpa2390,
Unless I’m misunderstanding you, I think the feature already exists. Its not exactly easy to find, but its there once you know to look for it.

If a device says that it is running, but it is actually not, you can tap on the device to open the details page. Then tap the gear in the upper right hand corner and then switch from Details to Manage. Once on the manage tab, scroll down to “Report a problem”. You will then be able to report “Device is not on”.

But even the “Device is not on” is only so helpful in some cases. I currently have a ticket with support as I have some lights that are “identified”, but something else triggers that device. So when Sense says that the lights are on, and I know they aren’t, I report “Not On”. But in doing some math, the light bulbs don’t add up to what Sense is reporting, so I’m thinking that the inrush of the light bulbs matches something else that runs, but I haven’t found it. So possibly saying that the lights are not on is actually doing harm as the “real” device is the one I don’t know yet, and the lights are the red herring.

But to get back to your suggestion, I do think the next step is adding a “device is on”. It would only work for devices that are already identified, but even if you could press “Device is in” and then it would ask" did the device turn on in the last 30 seconds or last 5 minutes" then the data science teams could in theory look to see why the turn on was missed. Or as an expansion of the “Device is Not On” feature, add the follow up question of “was something else confused as this” and you can select your humidifier. So again, the data teams can look at the logs and see that at xx:yy:zz.nn we thought that HVAC turned on but it was actually Humidifier.

1 Like

Paragraph 1: that’s exactly what I am looking for.

Paragraph 2: I don’t understand how that would be doing to harm to the real device. seems like sense would recognize that it is not the lights and then start trying to figure out what it actually is. So chances are I don’t understand what you mean here. before I start using it, could you clarify for me. and keep in mind that I am not overly familiar with this area of science.

Paragraph 3: yes. what an excellent addition to my suggestion. that is exactly it.


I just tried that. and it is a written report. nothing that should confuse the device. you send in a written report for someone to read then look at the data.


Hi Ben and Grandpa,

Two things:

  • I think you have a misunderstanding if you think Sense data science engineers look at the “log” every time we register a “Device Not On” tag via the “Report a Problem” menu. That tag goes into the dataset for your devices, but is probably only read during machine training to help identification and to better avoid misidentification. In a recent user session, the Sense data science team said that there was no way their team could deal with all the incoming feedback data in a humanly useful way. 99.9999% of the feedback data is only touched by Sense’s data analytics and machine learning.
  • I suspect we will see better ways of tagging conflated identification and for separating “twin identifications” where two devices initially appear the same to Sense. That would be the best addition to Sense’s end user feedback ‘ manual training.
closed #6

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.