INVESTIGATING: New device history not filling

I’m sure over a few days this will backfill, but alerting me before it has happened seems to offer little value. Until the history is there I cannot determine what it is to name it or categorize it.

5 Likes

A value I can identify:

  • Sense can assume that its detection is incorrect (which I believe would be the default state)

  • Meanwhile another detected device has been sitting in the list and could conceivably be the actual (or not) “Motor 1” in your case.

  • The detection and listing still gives you information: In your case motor.

  • The other information, even if you get an empty “Device X” listing, is that you have a device so it might give you pause on assumptions you make on other detections, which is a good thing.

There are probably more coherent arguments.

Having it show up now also lets you turn on notifications before the next time it runs.

1 Like

Valid point but still not my preference. :slight_smile:

1 Like

Eventually this issue speaks to the need for an indication of “Confidence” or “Detection state” in the UI.

e.g. Initial “Motor 1” screen could contain some verbiage including “Set a notification to aid detection?” vs. empty data … and potentially an indication of time since initial detection & confidence/state.

1 Like

So a question - How many detections do you want it to have before it shows you.
So in theory Sense has been watching the signals for a long time. It slowly starts to see a repetitive pattern, or a signal that looks like its “a thing”. At some point in time, it goes “ok, I feel really good this IS a thing” and identifies it.
How many more times does it have to see that thing before it tells you that it saw it, both forward and backwards? As far as I can tell backfilled data is relative to the device that is detected. I think there are certain devices that are clear enough and easy enough for the algorithm to go back in time and find cycles it missed. I think there are other devices that probably are harder to go back in time to find all the detections it missed. I’m sure Sense has logic in place for how far back their system looks, but I ask because I think the answer complicates the “wait until it has the backfilled data” request, and of course may differ user to user.

1 Like

I’d prefer enough detections and data so when I open the app I can look at it and say “Yes, that patterns aligns with a device of this type we have. I’m going to enable notifications and pay more attention so I can confidently name/categorize it.”

Admittedly if it is a device used sparingly then the threshold may never be hit. Perhaps there’s a middle ground where sense says to itself “I’ve waited X days after my confidence threshold was reached. I’ve tried to backfill or grab new data, yet the thing hasn’t shown up again during that time. Let’s make them aware anyway and maybe they’ll go turn it on for confirmation.”

1 Like

I wouldn’t think Sense would have to “go back and look”. It would have already been identifying these patterns and been keeping track of the data and history.

1 Like

I don’t believe its quite as simple as backfilling based on some current model because, I would assume, the model is a match with high resolution instantaneous data (at the monitor in realtime). That data (the high resolution stuff) doesn’t all get uploaded and stored in the cloud.

@kevin1 has no-doubt thought harder about this and there’s probably some waxing elsewhere but I would guess that the monitor’s internal storage is caching and that’s what’s used for the model match. If there are certain other “gross” (lower-res) models that can be applied to backfill then the confidence would no-doubt be less but could have the advantage of deeper processing on the mothership.

I think you guys are arguing for the same thing: not so much a mod to whether the data gets propagated before or after device identification as modifying the initial device view as I suggested above.

But I think thats kind of my question. When did Sense detect the device with enough confidence that it felt it could even call it a device, and what is that relationship towards historical data. The fact that historical data does not necessarily show up right away is what tells me that it has to go back in time a bit. Otherwise why wouldn’t it be there when the device is first presented to the user.
That said, to @kevin1 and @ixu point about where the detection is actually happening does bring to question the ability to look back in time.

@scorp508, I do agree with you about maybe there is something to a time out of displaying without history after x time, but I think it still all comes down to understanding when the system detects a device vs when it feels confident to tell you about it.

1 Like

I had a detection awhile back that I was talking to support about. While talking about it, they said they could see where it had been on the verge of being detected for some time (don’t remember the exact phrasing used).
This told me that it must happen in steps or near detections are in a queue of sorts before being presented. In order for that to be, the device would have to have been tracked.
The complete detection process is tight lipped so we have some guesswork as to how it works.

There is a tendency among those on the edges (us) to overthink these issues. I have strong confidence that there are higher minds at Sense working on things. That said, where I see our input being worthwhile is on just this kind of feedback … at least in the OP.

I just looped back to a thought I had elsewhere: Why is the Devices list limited to alphabetical? If you could sort it with AO (collapsed) at top/bottom and then in order of ON-ness (is that a word?). Most ON at top or bottom, selectable. I suspect it would go some of the way toward negating the issue in the OP … if you follow my drift.

1 Like

It’d be cool if there was a “possible devices” section that showed devices that Sense is monitoring and thinks you may have, but isn’t positive about yet. Doesn’t even have to give us anymore info than that.

1 Like

This used to exist and was removed.

The name of the game with everything is managing expectations. The biggest fault of the product is that many come in with higher expectations than the product can deliver. Sense has addressed some of these expectations and not others.

The “possibly devices” gave an expectation that the devices that were possible would turn into actual detections one day. The reality is that many of those possible detections did not come to fruition, or at least not in a timely manor which caused more questions.

Or at least that is my take away from why that feature was removed.

1 Like

Yeah was just about to come here to complain that it’s super annoying when sense has found a 3rd heat device and then immediately wants the user to pick whether it’s a microwave, coffee maker, toaster, etc (all of which have less than 20% confidence) yet the damned thing can’t show me when it ran or last ran. I dunno sense…YOU tell me since you’ve been monitoring things for me!

2 Likes

Quick heads up on this: I’m not ignoring this thread :slight_smile:

I’m talking with the Data Science team about why this happens. In short, it shouldn’t be and there is a bug filed with scheduled work soon.

Thank you @RyanAtSense,

‘Motor 1’ is still showing zero data from 2018/2019 and 2017 is pre-Sense. Team has perma-permission to look at my data.

2 Likes

Yeah I still days later see zero usage for this device. None in all of history.

Today for the first time ever I have 10 minutes of data on this device. Kinda funny. But I doubt it’s a new device it’s tracking.

You beat me, my Motor 1 still shows no instances of being on in 2018/2019.