It’s incredible how much worse my Arc experience has gotten over the years. Here’s a little snapshot of how it’s working now: I mark one segment “airplane” for the first time in months, and now everything I do is airplane and needs to be corrected. Moved 50 feet in 5 minutes? Airplane. Drove 10 miles at 60mph? Airplane. Stood around for 10 minutes then walked 3 blocks? Clearly you took an airplane. Airplanes all day every day, no matter the speed or distance.
Also, amidst spending a few hours in a theme park, here’s 50 different “place” entries that you need to confirm individually. Oh, and after you label 25 of the 50, the other 25 will still say “unknown place” even though they’re only 2 feet away from the other 25 places you already confirmed.
It’s just insanely inane and tedious. Been using Arc for years and it just seems to get increasingly obtuse in how it labels/categorizes things. You’d think the “learning engine” would have learned by now that 2000 miles at 500mph = airplane, 20 miles at 60mph = car, 5 blocks at 3mph = walking. But it just completely fails to understand this even after years of usage and confirming/correcting. It’s essentially unusable at this point because I’d have to spend 20-30 minutes every single day correcting every single little thing. I’m open to any advice or suggestions, but it’s frustrating as a paid lifetime user that this is what my experience has devolved to.
What you’re describing is not normal, and the result of something going wrong on your phone.
If you have been using Arc for years, what you describe here is not possible, given the massive amount of data going into the activity type models. Something is very wrong on your phone for that to happen.
Again, very much not normal. Something is wrong.
I suggest you step back and reassess the situation, and approach it more as a fault finding situation rather than an attack and complaint. Something is wrong specifically on your phone, and we can track down the cause and get things back to normal. Are you ready to start that process?
If you have been using Arc for years, what you describe here is not possible
Indeed, it seems almost impossibly bizarre. But to be clear, the issue has been gradually worsening over months/years, it didn’t just happen overnight.
Are you ready to start that process?
Absolutely! I’d appreciate your help rectifying this. Where do we start?
Ok, first step is to check whether you’re running Arc Mini or Arc Recorder as a companion to Arc Timeline app.
Recently a bunch of people have been finding that Arc Mini is causing weird problems with timeline data. Arc Mini is unofficially deprecated, replaced by Arc Recorder, and will eventually be removed from the App Store once I get time to document the change.
If you’re not using Arc Mini, then… I don’t have any good guesses unfortunately. What you describe with activity type classification is extremely unusual and not something I’ve seen reported at all, and I’m not aware of any weaknesses in the system that could cause that to happen. To I’m stumped on that one, even considering Arc Mini in the mix.
The second problem you describe, with many different visits to the same place, that does sound like something that could potentially be caused by the breakdown in communication between Arc Timeline and Arc Mini.
I don’t know why you are so confident that it’s not Arc’s problem.
I have lots of small problems including the situation the OG posts. Why am I not sending all these to you? Cause I don’t have time for every small bugs to debug or to be blamed as my phone’s problem.
For the problems I mentioned, they either categorize by you as other’s not using it (export gpx bug) or not havibg a result (arc mini stills shows over a year of not doing the task) or otger situation. For the problem I post, I never found a real solution.
Even if the phone has problem, I mostly cannpt deal with it cause it either will be restore as new phone or apple’s problem. So results in phone’s not normal will never workout.
I’ve introduced Arc to a lot people in past years. Lots of people has problem and quit to use it. Guess what? They will never give you such ‘I quit cause xxx’ feedback, because, most people just don’t give feedback, no reason,(as they don’t have the responsibility to give feedback either.)
I can bear with a lot bugs cause a software will always have bugs. But for over years, I don’t see Arc improve in any way. just normal with small develop. Thus, I don’t report anything anymore.
I have never used Arc Mini or Arc Recorder, just the main Arc app.
Have you ever considered adding a “reset learning engine” option, or a “reset activity type detection” option, that would retain my entire history up to this point, but essentially start the learning engine over, where it could begin learning my activity types all over again from scratch? I suspect some glitch happened in the past, deep in my older history, that’s gradually caused things to go off the rails with the engine. I’d love a way to retain my history but get a fresh slate in terms of activity type detection. And from browsing this forum and App Store reviews, I think I might not be the only person having this sort of issue, so I suspect other people could potentially benefit from having such an option as well.
I do know why. Because I have lived and breathed this app for the past decade, I deal with all the support issues public and private, and I have an excellent private beta team.
@o1x, please post new discussions for any separate issues you have. For this discussion, please keep it on topic to the original poster’s issues. If you have those issues and can add any extra useful context, please do so. If not, please don’t derail the discussion.
If you have general complaints with the progress or development of the app, please start a new discussion for that. This discussion here is for the problems that have been reported by the original poster.
Ok, curious. Then we’ve got some other weird problem(s) going on. We’ll have to dig deeper.
It wouldn’t serve any useful purpose. The classifiers learn from all of your explicit confirmations and corrections, and any “reset” would simply recreate the learned models from the same confirms and corrects all over again.
The only kind of glitch that could do that is explicit confirms or corrects that were mistakes and should have been something different. If you think there’s some old items incorrectly confirmed with the wrong type or place, it’ll make sense to go back through your data and look for those and fix them. That will take that bad data out of the classifier models and lead things back to the correct path.
That’s not the case. The current activity type classifier system is extremely solid, and the best it’s ever been. It’s based on Apple’s Core ML classifiers, which are excellent, and works great. There are no known problems with it. When there are problems with it, believe me I hear about it, repeatedly, in public and private. That is not the case with the current system.
There is something specific on your phone that is going wrong, and we need to track down what that is. It could be old data that’s been incorrectly confirmed, it could even be database corruption. But what it isn’t is any significant fault with the current classifiers system - that part is in excellent shape, having evolved and refined over many years, working with high accuracy levels for tens of thousands of users over centuries worth of accumulated data.
Let’s not let this fault finding process get derailed. We need to find clues, identify specific cases, share screenshots (you can email me examples at matt@bigpaua.com or DM them to me here if you don’t want to share them in public on the forum), identify common mistakes that are showing up in common locations / times / activity types, etc. We need to play Sherlock Holmes with the problem, tracking down its source.
I want to reiterate this point. When there are problems happening to more than one user, I hear about them, and I lose sleep over them. There are areas like that that do currently keep me awake at night (the complexity and unreliability of the backups/restore system is one).
There is no such situation like that with the current activity type classifiers or place classifiers. I sleep like a baby with regard to those systems. Which is why I am confident in saying that there is something specific with the situation on your phone that is very much abnormal, and we need to fault find it in terms of the data and app on your phone specifically.
So let’s continue doing that, so that we can get the problem on your phone fixed!