Hi! Love the app. One point of confusion for me, though, is why the app will seem like it’s resolved my day, account for my whereabouts, labeled my places and activities, and all is well…then when I open it again later, it starts thinking again and then shows me the uncertain items banner with a number that will sometimes change as I look at it. Why does it seem that my timeline is resolved, but then the app will decide there’s all kinds of things it needs to think more about? Will this get better with more usage? Thanks!
1 Like
Definitely!
Though there’s a few different things going on, so I’ll break it down.
The first thing is potentially changing number of unconfirmed / uncertain items. While the processing engine is doing its job it’s doing it iteratively. Meaning that it is looping over the items again and again, looking for the best possible merges and cleanups to do, doing the best one, then repeating that until there’s no more possible merges or cleanups left.
After it does each merge the timeline will have changed shape, so the list of potential merges will have changed, and the “merge scores” of each one might also have changed. And also how uncertain the classifier is about each item will also change. If two uncertain items are merged, that might result in a new single item that the classifier has high confidence in. (Or less likely, a new item that the classifier has lower confidence in).
That’s why you’ll sometimes see the number of uncertain/unconfirmed items change over a few seconds, as the processing engine is doing its job. Though typically that number should go down rather than up!
Ok, so the next thing is why sometimes when you come back to a previous day it will now say it’s got more uncertain or unconfirmed items than it previously did. There’s two reasons for that, and the first is simply that the latest Arc update now shows “unconfirmed” items, which it didn’t previously. These are items that the classifier is happy with - it thinks it made the right choice - but that you haven’t explicitly confirmed yourself. Offering that button for the unconfirmed items lets you explicitly confirm them, locking in their type or place assignments so they won’t be automatically changed later.
Because Arc didn’t previously surface those unconfirmed items like that, if you navigate back to earlier days that were recorded before this latest Arc update you’ll see that grey “unconfirmed” button appear, giving you quick access to confirming them if you want to.
The second reason why a previous day might show more “uncertain” (rather than “unconfirmed”) items than it did previously is that as the classifiers learn from new data, they will sometimes become less confident about their previous choices. An example will help to illustrate this.
Imagine you’ve just started using the app, and the only data it’s seen is you walking from home to the local convenience store and back, for a few days of the week. The classifiers learn from that, knowing that you walk that route. Then the next week you decide to cycle to the convenience store, and you correct that item in the timeline to be cycling after the classifier initially thought it was walking. Now the classifier learns from that, and realises that you don’t only walk that route, you sometimes also cycle it. Which means that its previously very confident walking assignments now look more questionable - it might have been cycling. So when reassessing those older trips it now realises it shouldn’t have such high confidence in its walking assignment anymore. Those previously certain items now become “uncertain”.
Eventually the classifier could gain high certainty again, if it learns things like you always walk to the convenience store in the mornings but cycle there in the afternoons. But as a general rule of thumb, the more variety of modes of transport you use, the more challenging it is for the classifier to feel confident in its choices. Some choices will still stand out as obvious and certain, but some might always be a challenge. For example if you travel a regular route, but with a mix of either car, bus, taxi, or cycling, then the classifier might still learn to get that right most of the time, but might never feel entirely confident in its choices, and will need to ask.
The last thing that might be going on is that “Thinking…” you see in the actual timeline itself. That is shown when there’s one or more very short/brief timeline items, that don’t meet the “worth keeping” threshold, and are likely to be merged in to an adjacent item once processing has finished. Once the processing engine has finished doing its cleanup, those “Thinking…” labels are meant to disappear, and any short/brief items that still couldn’t be merged will reappear in place of the “Thinking…”
But there’s an annoying little bug there, where sometimes even though processing has finished, that “Thinking…” box gets left in the timeline. Something in the UI hasn’t updated properly! So if you see that box stuck there for a long time, it’s probably just a visual glitch, and will disappear on its own later on. That data probably isn’t still being thought about.