Detection questions/requests: Self report partial detection, and better device categories

[surias.0]

A slightly rambling post with a few questions and suggestions.
I know that some of the assertions I’m making are likely greatly oversimplifying the context, and may be accounted for via other mechanisms such as the very large data sets of device detections you can compare against and iterate model training against.

• Can we be given a way to self report partial detection’s? I understand that the training of models is unsupervised and incredibly complex, but I have to imagine this could help if properly curated. How else could Sense know there is a partial detection? Just as the ‘Is this a guess’ toggle is important, the lack of allowing people to self report known partial detection could lead to false training/assumption of full appliance detection.

• Can there be a category distinction created between ‘Coffee Maker’ and ‘Espresso Machine’?

• In the Electronics category, would it not be practical to allow us to identify a ‘Projector Device’ instead of assuming that TV’s are the only display device?
  Would you consider training a model to identify UPS devices? This could be augmented by network detection for UPS’s that are plugged to the local LAN.

Partial detection examples:

1. I have a Breville 800ESXL/B Espresso machine. The only time it is detected is when the Steam Wand is being operated. Never when the machine is pulling an espresso shot, warming up, idling, etc.
   Can you give us some mechanism to report this sort of thing?
   Would it not be practical to make a distinction between Drip Coffee makers, Espresso Machines, and ‘capsule coffee makers’? A drip coffee maker boiling water must have a pretty consistant but different power signature than an espresso machine compressor boiling and forcing water for espresso shots and vaporizing water.

2. My Maytag Jenn-Air refrigrator/freezer combo is only partially detected. Less than half its regular power draw. I’ve provided Make, Model and given it a suitable device name. I have no way to advise Sense that this is not(or may not) be a full device detection, which would almost certainly lead to a false assumption that it ‘may be’ or ‘is’ a full device detection.

3. I have several Electrical Baseboard heaters throughout my home. I have several ‘heat’ devices detected, and have indicated in the Sense app that they are Baseboard and provided the Make but not the model. Why? Because these heat devices only seem to detect a fraction of a percent of usage, very infrequently, and only for 3-5 seconds at a time, despite the fact that the heaters on on constantly (its winter and cold outside in Eastern Canada).

[qrnef]

I like this idea, and I to add to what you already wrote, I think it would be great if a notification (like a new device notification) occurred when new signatures were added to an existing device (especially useful if we have something marked partial, so that we can revisit whether or not to “promote” it to a full detection). I suggest moving the tag to “Product Wishlist” so others can weigh in their support for something like this (and the product team uses community support for wishlist items in their prioritization of new features).

[chrisd253]

Great idea. I’ve only had mine hooked up a week and a half and it’s done pretty good finding things but most of them it’s only found part of them. Like my stove, it only recognizes one burner out of the four as the stove the other three in the oven belong to other still. Also three out of the four burners are double burners so they have two elements so who knows if they’re getting both of them

[ixu]

Logical. Though I imagine Sense has thought carefully about this and put it on the low priority list. Here’s my guess why:

• Your #3 example points to why user input could often be misleading and the weighting pointless. With multiple similar devices, like similar baseboard heaters, you’ll get lots of on/off signatures and Sense has very limited ways to distinguish which heater is going on/off. When you add in possibly-incorrect user input you may be adding to the problem.

• Suppose there is a distinct (unique) device that you know is not getting complete detection … Sense still needs to be able to track the overall electrical signature (not just the partial on/off and tracked usage). The reason that the detection is partial, meanwhile, is likely for the reason above – Sense is confusing the ons/offs with other devices. So we’re back to same issue.

In some cases it would surely help but it’s a matter of weighing that effort (and data) with tweaks that improve unsupervised (more optimal) disaggregation.