So there is a Mall.
First time I come to the mall I’m going to Place A and so I marked it as Place A.
And overtime I most never go to Place A again, but it was always shows default place as I come to that place.
I think the algorithm should marked the mostly choose place name as the default one, rather than the first time place.
Are you confirming the visits to the “real” place (Place B)? If you do so, Arc should learn pretty quickly to prefer Place B over Place A. To check, open the Place B visits and check if it’s requesting confirmation (green button). How many visits do you have with Place A and B?
What I did:
- when confirmed it shows Place A.
- I scroll to the list and found Place B and then it changed.
If above means confirm. I’ve already did at least 10days with more than 20+ confirmation.
There’s several factors that go into the classifier’s decision for scoring place matches:
- Distance between visit centre and Place centre
- Amount of overlap between Visit’s coordinates matrix and Place’s coords matrix
- Time since last visit to that Place (this is the one you’re suggesting)
- Time of day of the visit
Also if it’s not the current visit, there’s one more:
- Similarity between the Visit’s duration and the Place’s common visit durations
Then finally, the classifier weights each place score by the total number of visits to that Place - ie a place with 100 previous visits will be higher ranked than a place with 10 previous visits.
You can see most of this code in Place.scores.swift in Arc Mini’s code. It’s essentially the same code as in Arc v3.
So what’s likely happening in this case @o1x is that Place A has had a lot more visits, so even though Place B has been visited more recently, it’s still scoring slightly lower, due to Place A having had more total visits.
I’ve had a similar problem, where my local mall used to have a Tom n Toms cafe, that I came to every day, but it’s now been replaced by a Starbucks, which I use instead. The Tom n Toms has 850 visits, but the Starbucks only 150, and even though I haven’t been to the Tom n Toms in many months (it stopped existing maybe 6 months ago), it’s still sometimes the top match (with Starbucks coming in second).
It’s probably worth me doing some tweaking on the weighting of those factors. It seems a bit troublesome for places that you might’ve gone to a whole lot, but that you now never go to. Perhaps stronger weighting on the “time since last visit” factor would do the trick.
Extra thought: It’s also much worse when Place A and Place B are essentially in the same location, for example in my case where Starbucks literally moved into the space that Tom n Toms moved out of, so the location data is identical.
The classifier typically does very well at recognising the distinct places as long as they’re not in exactly the same location. Though if they’re literally on top of each other (eg on different floors of the mall, but in the same corner), that’ll confuse the classifier too.
let me try more days and try to confirmed more days to see if it’s changed.
The wired thing is Place A only going to about 4 times with 1hour each time. But place b is already lots of days and more times
That is weird. Yeah, see how it goes over a few more visits. Though with Place A only having 4 visits, it surely shouldn’t be preferred by the classifier. Unless perhaps its most common visit time was an exact match for the mistaken visit, and the centre distance was an exact match too. Maybe just a freak once off thing.