Detection has devolved into a mess with no support from Sense

technicalsupport
device-detection
#1

At some point, our Sense decided to regularly attribute usage to our paper shredder. I fairly regularly go through the workflow and report that the device is not on. Since we’ve now unplugged it entirely and it’s still attributing usage to it, I filed a regular report with text. No reply from Sense support.

I don’t understand why I’ve not heard back from Sense with at least some specific advice like deleting the device or resetting everything. Are they still in business?

It’s been over a year and our fridge, freezer, induction cooktop, EV, toaster, and many other large draws with clear patterns have still not been recognized. It’s also forgotten about our sink garbage disposal and added a second one which it now attributes use to.

#2

Hey @alexr - very sorry for the troubles you’re experiencing!

Can you PM me the details of when you submitted your support request so we can follow up and make sure it didn’t get lost somewhere. We’re very much in business and our support team works hard to try and respond to all tickets within 24 hours, so we absolutely want to get this figured out for you.

#3

Two thoughts…

  1. I used to be fairly OCD over watching everything my Sense did and tried to provide immediate feedback every time I saw a mis-identification, even though I know enough about machine learning to know that this will be a long learning curve.

  2. Sense folks do answer support emails and via forum. The general advice I have seen is to NOT reset everything, and to continue to let things play out without deleting devices, while providing Sense feedback when a well-understood device is off. One interesting thing I have found is that some of my detected devices are “seasonal”. My AC units appear as one set of devices mid-summer, but a different set in the fall and spring.
    Sense vs. Ecobee

There’s a lot to learn about disaggregation of electrical usage and Sense keeps progressing, but facial recognition was an 8-10 year learning curve with machine learning, so I now take solace in a slightly altered version of one of Dan Fogelberg’s songs…

Sense has these moments
All steady and strong
Device recognition rumbles
The next thing I know
Sense is guessing and lost
And all recognition crumbles

The meanings get lost
And the teachings get tossed
And you don’t know what
It’s going to do next
You wait for support
But without a retort
Some kind of message
Comes through to you
Some kind of message comes through

And it says to you. . .

Feedback when you can
Reset when you have to
Do what you must
That’s a part of the plan
Await that detection
Without feeling rejection
And one day we’ll all understand …

Sung to the tune of…

1 Like
#4

I appreciate your points, however, I’ve been working with ML/CNNs for a long time now (1994 or so). Where this is failing is humans.

1 Like
#5

Glad to hear about your long history with ML and CCNs. My experience is that they have only become consumer-worthy for some of the earliest applications over the past few years. Have you investigated the MIT, CMU and Berkeley work with electrical disaggregation ? I haven’t seen anything close to what Sense is offering.

FYI - I lived through the issue-filled and corruption-prone messes of both Google’s and Apple’s premature rollout of facial recognition. There, a huge investment in tagging was often rewarded with a complete loss of data or by by hyper-aggressive (and incorrect) labeling of faces. And it took about 3 years before I saw software from both of them that was truly fit for consumer usage. I kind of prefer guys on the other end doing most of the work, as they are with Sense.