Both can be true. Check out this video to learn more about “noisy” devices. George, our VP of Tech, goes into it pretty in depth.
I am very new to sense, I have only had my device connected for about 2 months so I am sure I am missing lots of important information.
But it seems to me that if the Sense software would just let the user enter a list of the specific devices he had in the home it would be very helpful to the identification algorithms.
For example if I told the Sense software I had two Air conditioner units that were of a specific brand and size, that I had a Tesla with the specific charger, an LG washing machine model XXXXXX, and a LG dryer model XXXXXX, and 6 watt LED lights in the bedroom and 15 watt LED lights in the kitchen and 3 TV of Type XXX, and 2 computers of type XXXX etc. Then the software could immediately start using the specific models of those devices to look for them. It just seems that the more clues the software has the better and faster it will be able to identify loads.
Why am I wrong?
Look in Settings > My Home > Home Details > Device Inventory.
You should like what you find.
This is a new feature (as of last updates) so will evolve. If you have any thoughts or comments after playing around with the Home Details and making your Inventory, the (lively) discussion has been mostly here:
Thank you, that is exactly what I was talking about.
The only thing about this new feature is I believe it’s just collecting data for now. It won’t help Sense detect the devices in your home just yet.
It’s a great step in the right direction.
Agreed, though to invert your assertion I would say that helping now with data input (making a list) will help Sense “now” with “future” detection if only because it may influence the ML coders’ tweaking. That goes to the core of the Sense challenge. TOTO!
Yes @ixu, I believe that completely and feel that I’ve already benefited on my detections from the users before me entering their models on the individual device details. This new feature is just a better incarnation of of those details.
I’m guilty right now if not having all of mine filled in completely. Something I definitely to get caught up.
This is an oft-asked question and I thought I’d give you my speculation on why this intuition that knowing the devices would “help” with identification isn’t true (at the moment).
I wonder if it is more of a case of asking, “why would such information be valuable to a computer algorithm?” I suspect the answer is that people think such information would be valuable because of how people would approach the problem of device identification. But the algorithm isn’t a person and doesn’t approach the problem like a human would. It doesn’t have the limitation of only being able to check against a limited number of device signatures at a time, so it is probably always “checking” transition signatures against all known devices and all known classes of devices.And this is in practical terms no harder to do than to only check against a subset of them.
So we could tell Sense what devices we have, but it will still check against all devices in the known universe, and beyond (I’m looking at you, Light 1).
The problem with crowd sourced info is that three people might enter the same device model three different ways. This puts the merging of all crowd sourced models together into one for all so that the same model is tracked and detected the same across the platform.
Generically, I’d just be happy if it detected a model and kept detecting it. I’ve been around the circle repeated now with my HVAC. First it is detected as “motor1” and “ac”. Within a day or “motor1” never shows running and when “ac” is on I will see the other side of it showing up under “other”. My hvac is 240volt and the air handler is 120v.
I’ve also had two different “dryers” identified but only have one. My washer is detected at times. I set a rule to tell me if it stops running for 5 minutes so I know the load is done but if it doesn’t detect the start and it’s running under “other” that’s pretty worthless. I’m sure this same thing would occur for the gentlemen who wanted to know if his garage door changed positions by knowing if it ran.
My user experience doesn’t really line up with the advertising (yet). Aside from knowing over all ct’d usage and chasing phantom loads, I’m not really sure any of the device detection is both reliable or useful yet because it hasn’t been reliable. I mean, hell, even my device specific usage totals are worthless if it misses them run frequently. So how good is all that user days presented to me.
After watching the founders speak on the “ask sense” videos one thing stands out to me. Each month they get more units sold and more inventory hanging off of them to attempt to discover and model/detect. We are in fact beta testers. Will this ever improve beyond that? I’m not sure. Device detection is hard. I get it. They are dealing with 20+ years of hardware out there and wiring issues to boot.
I agree with you @dottat.mobile all the way. We are beta testers. I see this constantly on the form, how patient do you have to be before things improve. There is something wrong with the algorithms they are using but apparently no one is willing to change anything. The proof of that is nothing changes. I also have an AC with unusable data. Three different wattages and sometimes it is on “Other”. Now I read a lot of chatter about noise being the problem, if so how do you clean that up. What next?
We get the “why can’t I teach it” question quite a bit. Check out this thread to learn more: Why can't you train Sense?
I didn’t ask that. My concerns are more along the lines of it being a circular pattern. Round and round and round. Finding and forgetting. Repeatedly. Making all the data pretty useless.
You know the webpage advertises stuff like “being able to tell when the kids come home”. I just wish you would relay to your marketing people they are way ahead of the reliability and current capabilities of the product. Way ahead.
I’ve also had devices detected that will not show as that detection but another at times. It can depend on the cycle I’m running. Take my washing machine. The detection I named “washing machine” comes up on normal and heavy cycles but when I run delicates, it uses a different load and it’s my “device 1”.
For a long time, a delicate load just went to other so it’s nice to have this new detection for certain loads.
For many people, this is very doable, between detected garage doors, lights, televisions, toasters, water heaters and so on. But as I’ve noted elsewhere, and as we note on our site in multiple places, the device detection experience is not universal and can vary across homes for a bunch of different reasons. In my home (and I get no special treatment), I know when my wife gets home from work based on the fridge opening and the lights coming on.
I’m going to work with the Data Science team to flesh out some better answers to questions around inconsistent detection. In the meantime, the bottom portion of this blog addresses it to a certain extent, particularly @samwooly1’s point on device modes.
Ok so you must be either very lucky or well taken care of by support. To the point you can detect lights in your fridge.
It’s the round and round cycles of detecting and forgetting or detecting and then redetecting the same device again as an entirely different device. Or detecting full wattage of a 240v HVAC compressor and later only detecting one of the legs instead of both (either dumping the remaining load in other or into another device).
If you had that going on at your house you’d likely crawl up supports you know what until they sorted it out for you. Everyone can’t do that.
I understand why you feel this is a bit overstated, I agree to an extent. But I also can tell exactly who is and is not home just about anytime. It’s the patterns that I personally recognize by length of showers, fridge door opening and coffee maker being used. Those are just a few of the indicators in our situation. These are things I never paid attention to in the past.
I guess what I’m trying to say is I dont need them focused on that. I have home security and cameras. I just want to know my devices and what they are using with reliability it currently does not have.
I don’t think I had any support prior to the fridge components detections. My fridge is detected as the compressor and then separate detections for the water/ice door dispenser, fridge door light and defrost cycle.
The defrost doesn’t work quite right. It detects the On signature but not the off. So the defrost appears to run longer than it actually does.
Also, I don’t think it’s within supports capabilities to magically force detections. When I have contacted them about detections, they have been able to see the same things I have and stated that Sense didn’t have enough data built yet or there was something else preventing detection, like noise.
This is true. If they could force detections, I’d be bribing them with cakes and other treats both to help me out and to help you guys out. Alas, it’s not that simple.
As promised, I worked with the Data Science team to write up an article on the reasons why Sense may lose detections. Check it out and let me know if you have further questions.