Discussion: Why can't I train my sense?


I know. That’s what I was saying. and that’s not a problem for me. If Sense detects the device 10 minutes after it turns on (depending on the device), in my case, I’m fine. Most devices, it isn’t important that I know immediately whether it is on. Though it is nice to have a better response time than 10 minutes :wink:
the problem is when I’m looking back at devices and seeing that sense didn’t record the entirety of it. And I’m not talking about machines, I’m talking about individual devices. like when it didn’t start capturing my dehumidifier until it had been running for 3 minutes. and then recorded it as off 10 minutes before it did. (just making up numbers here). I should have chose a better example than the dehumidifier because it is a machine. but you know what I’m trying to say.


That was exactly what i was sting @Grandpa2390, sense is no match the human brain. This had been proven many times over and it will remain that way for a long time to come even the best facial facial recognition is no match for an Elementary age child.
A good article here:


The top-down server knows the best approach is not working for many users.
My “Other” bubble is large and has not changed for a week.
Expecting users to buy and then wait weeks ior months for your product to work is not acceptable.
Most users that buy a device like this know what is happening with loads in their house.
Add a way for users to identify loads in the app.

Device Detection Update: Device Updates!

I suggest reading through some of this thread as well as this post: Why can't you train Sense?. There’s a lot of detail here about why what you describe is not a feasible option.

Have you only had Sense for a week? That’s quite a short period of time for Sense to learn about your home. Machine learning requires a lot of data and Sense can only get that by seeing repeated on/off cycles of devices in your home. A few cycles just won’t cut it.


I did not say I had it for a week.
I said it hasn’t updated any loads in a week.


Being that sense is a fairly new company and with more people buying it I am sure that we are going to be seeing many improvements with the app and with more people using it I am sure that device detection will only be happening faster. I am looking forward to them adding the TP-link HS300 to the list of supported devices which will end up giving them more data for their server.


It looks like you’ve been a Sense user for a little over 2 weeks?

As Ryan suggests, even at the 2 week period it is very early in the Sense initial detection phase. I saw one device in the first week and a half or so (my furnace fan which cycles several times every hour, so it was an easy find) and then it was (IIRC) around the 2 week period where I started to see some additional devices.

Has your unit cleared the initial setup stages and is now indicating it’s monitoring?


The device is finding new leads but is often wrong on the type of load.
It has found my refrigerator and its light!

My freezer was also detected.

I am learning how to use the app better.

Some progress has been made but at a plodding pace.


The very nature of the way that Sense works means…the impatient are apt to be somewhat frustrated in the beginning.

I am like you, very…very impatient, especially with tech. When I first got my Sense unit and it took 8 or 10 hours to calibrate before it even started looking for devices, and then spent another 3-5 days with no devices detected, it was incredibly tedious for me as well.

That said, I came here to the forum and started reading up on exactly why this was the way things were happening and soon started to understand and appreciate how things were working. It wasn’t easy, but I relaxed and accepted that patience was going to be a virtue with it came to Sense, and many months later I’m generally quite happy.

So relax, and enjoy the ride - you’re still at the very beginning. :wink:


It’s really common for sense to misidentify a device. Motors identified as heaters or lights as motors. Sense has done what it is designed to do by detecting the device but needs your help with identification. Sometimes it takes some work to properly identify a device but having it be detected is something to be happy about.
Your frustrations will change to “I’m not getting detections” at some point. Sense is a bit of an up and down ride. You might have a period of many accurate detections followed by times where the detections are off or completely absent.
As @MachoDrone stated above and in many other posts, Patience is key.


Hi There,
First post in the community - so hi folks! :slight_smile:

Honestly not really blown away by the sense yet, its detected the fridge (and only when its pump is on, not on standby, so I can’t set an alert if the fridge is actually off when an alert would actually be useful).

I have read numerous posts asking for the ability to help sense detect devices, and the response from the Sense folks don’t seem to really close the issue for me.

Right now the device is looking at delta change, power signals and using an AI model to process these signals. There is no reason why there wouldnt be the ability to teach sense your devices. An Analogy of people in a photograph was used here but honestly that doesn’t hold. identifying a face is actually VERY easy, so not overly comparative to this.
My question is if you had exactly the same noise (background) and you switched a device on and off, only that device, while Sense was in ‘listening mode’. It should be able to get a better idea of that device. Then switch a different set of devices on and repeat the on & off of the target device - the AI at this point should be identifying the common signature across (this could be repeated on a different day, and different time - as each time the single device is switched on & off the AI model should be removing anything else that isn’t the same).

In the photo of people scenario it would be like looking for Mike in a crowd, but the crowd keeps changing, and eventually when there are enough different crowds Mike would be the only common denominator (couple that with the EXTENSIVE data Sense is providing their data scientists) i’m struggling to see why this is not possible.

The complicated part should really be the always on - as by definition these devices contribute to the noise and there isn’t the ability to disambiguate so readily (without a listening mode exercise) OR devices that are only switched on rarely as requires a lot more data samples during usage to single out.


Sorry to hear your not yet impressed with sense and the detections @jamieeburgess.
For your fridge, sense is working exactly as designed but there is a very simple way for you to setup the alert you are seeking. Go to your device page in the app. Select your fridge and navigate to the top right hand corner so you can open the management page. Create a “custom notification” and have it let you know something like fridge is “off” for “4 hours”. You may have to tweak the time on it according to how often your fridge normally cycles.
To the rest of your post. I’m in agreement with a lot of it and I’ve also been very critical of what I believe I was led to expect and what the difference is in real world use.
We are not the only ones as you will see by reading the huge number of posts repeating the same complaints. I’ve just come to except it for what it is. Having patience is difficult but really is key.
I’ve only had sense since January 15th of this year so about 2 months. I’m at 21 active detections now and all of them are pretty accurate. I’ve had probably at least 20 other detections that sense has removed or combined with other devices or I have deleted due to not being satisfied for some reason.
Your really early into the sense experience, it gets better with time
Hang in there and try to be patient.
You’ll soon get more detections.


I know what you’re saying and i also don’t like the comparison of face recognition either because they just aren’t the same. A much better comparison would be comparing it to radar detection since they are both analyzing energy wave forms and this technology has been around for decades as you can see from just two examples below. anyways like @samwooly1 said it gets better the longer you have Sense installed and they seem to be making great improvements at Sense and I am extremely happy with my little orange box.



@jamieeburgess, @mike_gessner,
You’re right - I may need to revise my factial recognition analogy a bit, because @jamieeburgess is right - Facial recognition really doesn’t need the full-on flexibility of machine learning for identification, nor does radar detection. But they both use the same two part process as Sense does - Identification followed by classification.

Facial identification is typically done using biometric algorithms that discover and measure facial features, or principle component analysis that translates facial reference points into simpler identifying metrics. @mike_gessner, I’m fairly certain that radar identification is done via DSP and convolutional algorithms, since most military systems were developed/designed well before machine learning really came into being. Both facial recognition and radar identification use fixed algorithms and a smaller set of well defined features to do identification than typicallly used for machine learning.

The real analogy should be the current state of the art in machine vision and image recognition, that uses complex neural networks like ResNet, to identify thousands of different everyday objects, animals, and other nouns.