We should be able to TRAIN this!

I have about 240 watts Always on. I identified them best by shutting down power on a breaker and seeing the difference in watts in Sense. This basically tells me the power associated with the breaker, and identifying that device(s) is much easier. Most of that phantom power is my cable modem and router.

And for those who don’t know what sockets are associated with what breaker, it’s a good thing to do in a household. Some people mark the plate with stickers with the breaker number on them. I just have a spreadsheet listing the socket or device (like fans and lights) and location. I have sometimes moved my devices between 20 amp and 15 amp circuits to better balance the load, as well as to the different phases of 110V (useful for standby generators)

Sense doesn’t identify devices by make/model. It identifies by general category models (heating elements, refrigeration, general motors, incandescent lighting, florescent lighting, etc), then used crowd sourced data to guess the most likely more specific device subcategories (Refrigerator, Wine Refrigerator, Freezer). You can then tag by selecting the most valid option. But neither you, nor Sense in many cases will know the exact model or set of models (since many devices are composites) living within model A of maker B. Plus you could have composite devices or hidden devices in your house that might present as something you don’t think you have in your house. So I can’t see reliably removing false positive either.

Yes, it doesn’t know the make and model (except when it does - there are items that have very specific signatures like Tesla EVSE). But if you have a refrigerator newly detected, when you already have all your refrigerators detected (let’s say the compressor part for argument), another refrigerator doesn’t make any “sense”. Hence, either one of the refrigerators already detected is wrong, or Sense has misinterpreted it as one. I guarantee everyone knows how many washers, dryers, refrigerators, dishwashers, stoves, microwaves, toaster ovens, garbage disposals, TV’s, overhead fans, garage door openers, EVSEs, heaters, air conditioners (central or windows) etc that they have, and types (e.g. wine refrigerator vs full size).

Heck, there have people with sense that have reported having more refrigerators than they own! If extra ones are detected, this is essentially telling data scientists that one of the refrigerator models is wrong, and is worth another look.

As for the make & model information, since the Sense people pour through schematics trying to understand what they are seeing in the data (e.g. in-rush current limiter of a TV turning on), knowing the make and models would allow them to validate the Sense data of your home to the detected types by their models and correct as needed.

A curiosity question:

We are able to report “Device is not on” to Sense under Report a Problem for all our found Devices.

Why are we not able to report Device is On?

(Clarification Edit): I’m not asking about training for un-found devices. I’m only asking about the found devices that have turned on, but for some reason Sense missed them this time.

This isn’t an ask…just a question about the double standard.

A good analogy might be facial recognition. There are really two steps to the process. First machine learning identifies the bounding boxes/ circles for all the faces in a photo(s). After that a human can tag the faces with specific names. Subsequently, machine learning can begin associating names with at least some photos. But until machine learning has identified the face as a face, there’s no value in telling it that Jack is somewhere in the photo. Some photo environments do let a user define a facial region and assign a name, but that data is entirely for human benefit and NOT used for learning, because there is no “identification” for machine learning to tag with the name. But you certainly can erase incorrect names that have been automatically associated with an identified face, form improved learning. Similar to marking a device as “not on”.

Now this analogy is not entirely perfect:

  • Sense “sees” only one single unbroken timeline from a number of different “dimensions” or features vs. a collection of discrete photos. And that timeline endlessly grows, though I believe Sense might occasionly break up the timeline as they add new detected features to their dataset after some of their firmware upgrades.
  • The current objects that Sense detects today are very short, likely less than a second, so you can’t even see the waveforms associated with an on event or an off event in the power meter view.
  • Sense has to do two successful identifications to reliably monitor power of a device, spotting both the on-event and the off-event.

Just created a pinned thread that should answer this question pretty thoroughly:

1 Like

Thanks to all at Sense with their immense patience regarding the many “new users” and “old users” who are less than 100% enamored of the device discovery process.
This is not meant in an inflammatory or hostile way: Given the quantity of Sense users I see up here who routinely find their expectations do not match reality with the current device discovery process, I have to wonder whether the COO or the CEO is an engineer or a “bidness person”.
I suspect an engineer, :slight_smile: for my view is that an engineer would “stick to his guns” as y’all have, NO matter how many “non-engineers” criticize the process and express distress over their anemic progress with the product.
A bidness person in charge, on the other hand, would see the enormous dis-satisfaction being expressed as only the tip of a huge iceberg of potential bad reviews and seek to mitigate corporate injury by adopting some semblance of engagement with the customers that want to contribute, facilitate, and expedite the device identification process.
I am unwilling to suspend disbelief regarding the position taken by Sense of the total uselessness/impossibility of user input in device detection methodology. Get these people inside the tent, pssing out, instead of keeping them OUT, pssing on the tent. (LBJ) :slight_smile:
I am concerned Sense can’t survive this level of discord which is only increasing with regularity. I selfishly WANT Sense to succeed, and for that it has to accelerate and concretize device recognition/detection, and provide some way for those that have gone many months with no progress to engage better.

Your guess is correct re: our co-founders. You can read their profiles here: https://sense.com/about.html#leadership

We are absolutely not ruling out user intervention in the process. But at this time, training is just not feasible. We see those development efforts better focused on ML-based device detection (which has seen improvements and there are many posts here by users happy with new device detection models we’ve pushed out) and integrations with connected devices that can give us ground truth data. User input, in the meantime, can be focused on giving us the contextualizing data to improve device detection in both of these camps — from reporting problems/device is not on, to surveys of the devices in your homes, to utilizing Community Names and renaming your devices, to turning on NDI and enabling integrations. All of that user input has proven hugely beneficial and we’ll continue collecting that data and more. Maybe as this tech improves, a training mode will be feasible. This is all still very young tech, so we’re definitely not ruling that out for the future.



First off, I want to agree with you. Customer engagement by Sense is crucial at this stage in the game, maybe via quarterly discussions / briefings under NDA, to educate, provide roadmaps and answer questions.

A “bidness” guy should take key lessons out the “Crossing the Chasm” series of books that dial in on the rollout of new technology products. This one illustration from the first book, “Crossing the Chasm: Marketing and Selling High-Tech Products to Mainstream”, explains a lot.

I believe Sense is hitting the “chasm” prematurely since much of their marketing originally targeted mainstream consumer customers. Innovator and adopter type customers are relatively easy to please with new products since they appreciate the innovation and are willing to roll with the punches as the product evolves. But for mainstream customers, who expect everything to pretty much work out of the box, Sense has a mixed track record - in some households it works reasonably well and in others, not so much.

Some of the lessons that the Chasm series books give for success are:

  • target the appropriate customers (don’t target the mainstream too soon)
  • know the key buying needs of different pools of prospects and existing customers (EVs, data downloads, HVAC, pricing models that match utilities, etc.)
  • knock off solutions in each of those areas with transparency and clarity.

BTW - I happen to believe that Sense is doing the last two reasonably well, but could still improve.


Quarterly customer webinars are definitely where we want to end up. I’ll be announcing another soon (my first time as host) for some time in August.

You definitely have a point about earlier marketing efforts. We’re really working to make revisions to our language and I’m here to make sure the proper messaging is getting across to our community (here and across our social platforms).

(I didn’t immediately recognize you with the new avatar!)


Looking forward to the next one… I’m hoping that many of the more outspoken advocates with strong wish lists and ideas on what they want out Sense attend.

And yeah, I decided to actually use an avatar, but a move from a basic “K” to a Big Red “C” is a probably little confusing…

1 Like

Nothing important to add here but Ag '98 and CVM '04.

1 Like

@NJHaley, that’s important :wink:

Thanks very much. I have had a long history, since around 1980, in technological development and beta service/testing around the PC and starting in 1996, with JDS’ Stargate implementation. I have served a variety of companies in both software and hardware, such as for US Robotics’ modems and Citrix’ serial protocols for thin client development (you know it well nowadays, but it once was nearly swallowed up by MS with the intent to (probably) kill it. So anyway, I would be willing to contribute in your beta program if you think I might be of use. I’m searchable to a small degree as “TexARC” & “ACOfTexas”, and my site arcarmichael.com provides some background. ?

Mine’s been running a days and finally picked up two devices, thought more would have been picked up. Not into electronics, but after some reading I understand the challenges. Having done systems design and some IE, we might need to take a different view of the issue. Sense is trying collect info to identify a device and we want to see a device bubble. I suggest there’s a “Identify” mode for the monitor panel that could be turned on/off. The monitor is showing +/- watts detecting changes. I can go to different devices one at a time, turn on the device, Sense detects and records the info, now add a popup that ask what was turned on, keeping it as a tagABC thru the analysis, I let the device run a bit, then turn it off, popup asking did you turn off tagABC or what did you turn off, that would stay with the info. Do a few devices, set identify off. Doesn’t mean tagABC will be in a bubble every time it’s on, but would eventually surface. I could repeat the ident process, just to tag additional records.

Just to add, I won’t have an issue of several “identify” sessions to tag the recorded signatures presuming over time Sense would analyze and confirm the device. Maybe needs to be another post, but related is identifying duplicates: 2 frigs, 2 a/c, multiple heaters and TVs??

I’ve got to believe that it would be beneficial for me to be able to manually tell Sense about loads that I already know, versus waiting for the ML algorithms to identify them. We should also be able to enter a training mode where I can manually switch loads on and off to manually train Sense. I see a couple other requests for the same thing, and agree with them 100%.

Along the same lines, I have a pump installed. I know how many watts it makes when it runs. that seems like obvious information to enter. If I fill out enough such devices it should be useful in figuring out what to look for.

1 Like

Echo this suggestion! This is marketed as a crowd sourcing product right?

Please read this thread. It discusses why it is not a feasible inclusion.


If we felt that manual training would a good use of your time, we would work on implementing it

Wow, that’s a tad bit condescending to say, especially when customers are declaring it to be worth their time.