Feature suggestion

Love the Sense device and the information its giving me about what is going on at home. I have had the device now for 2 months and it’s only figured out 4 devices and what they are.

It would be nice to have another device that Sense would sell that you could plug a device into it (one at a time) and power it on. This device would register the devices pattern / signature and relay it back to sense. where you would then be able to tag it / label and category it. Since it’s only one device at a time you know what it is where its and any other important information.

1 Like

Hi. Welcome.
Your suggestion of collecting data is called a Ground Truth Data.
Sense is actively working on improving detection. Devices like this are on their mind, and they may be preparing for a method just as you described.
The forum has many discussions on this. But unless you know what it’s called, I understand why it’s hard to seek out.
Search for Ground Truth in this forum, you’ll find previous responses from Ryan and other Sense employees.
I’ve suggested SONOFF devices for their open firmware / custom firmware capabilities and SONOFF is very affordable.
I’m including @RyanAtSense .

And for those who really want to understand what’s likely happening in the Sense monitor and back at the mothership, here’s a good research paper:

http://redd.csail.mit.edu/kolter-kddsust11.pdf

Hey @rabramson_98 - welcome to the forum!

And big thanks to @MachoDrone and @kevin1 for the helpful explanations and links. We do have some exciting integrations in the works that we hope will address the type of use case you describe. We’ll be sure to update the thread when more information becomes available.

2 Likes

Welcome, @rabramson_98. I won’t re-cover the territory that Brad has already tread, but for the future, we do have a ‘Product Wishlist’ subforum. Just be sure to read the stickied threads at the top: https://community.sense.com/c/product-wishlist

I agree that something needs to happen to aid device detection, even if it was an option that the geek-set could enable that would “confirm” devices drastically earlier than might otherwise normally occur, even if Sense isn’t really anywhere near sure exactly what that device may be. I’m struggling to understand how some devices with (having observed them on the portal) seemingly blatantly obvious and repeatable signatures (IE, things like our coffee maker, refrigerator, chest freezer, bar fridge, etc) have days later, even after tens or hundreds of cycles, still remain undetected. Perhaps via an option like this, earlier “guesses” could then be further refined and confirmed by the end users who choose to be a little more involved in device detection if they wanted to be.

I know that the current Sense algorithms stress that they want to be basically 100% sure that a device is actually identified before showing it, but personally, I’d be willing to have devices show up much earlier as ‘wild guesses’ and then let me do the legwork to track down what they may be and add further confirmation.

I just got my sense installed on Tuesday so I know I’m premature to be complaining about the glacial pace of device detection so far, but I can’t say I’m not, well…surprised at how it’s still seemingly skipping over what seems like obvious stuff.

Yes, I’ve read the FAQ’s an I understand I’m almost certainly over-simplifying it in my head, especially when it comes to more complicated signatures like televisions, etc, so I’m saying all this with a grain of salt as a complete newbie,

The most successful recent research in electrical disaggregation (that’s the fancy name for what Sense does), uses power probes plugged into every appliance in the household as well as a device like Sense to watch the waveforms at the mains. Those research projects have much better training data since they have full visibility into exactly when each and every device turns on and off, plus how much power each one uses while active. Even with all that additional training information, these projects are correct 75% of the time, at best. And that’s for the same exact house each time (same device in a different house is much harder to identify), plus that is immediately after training (device signatures and environmental factors change over time).

Sense is unlikely to improve via human training because humans are extremely poor at providing the kind of feedback machine learning needs to actually learn. But, I think Sense could benefit from connecting to power probes and other devices that could supply the right kind of feedback to train and adapt to drift over time.

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