> I assumed confidence to be the probability of each task. However, often they do not add up to 100. If they do not represent probabilities what do they represent?
The probability of each being true independently?
> How can Google possibly predict activity type between IN_TWO_WHEELER_VEHICLE vs IN_FOUR_WHEELER_VEHICLE ?!
Two wheel vehicles can often make it through traffic, in between cars.
Measuring how much the phone rolled back and forth could be a factor, too. You don’t have to keep a car balanced, you do have to keep a bike/scooter/motorcycle balanced.
I going to bet there are more bike driving people with Android phones in places like Thailand, Indonesia, India and the Philippines where on-road behavior is more similar to each other than to the US.
I think it's a pretty hard problem, and we shouldn't be surprised if the accuracy there is low.
It's probably a combination of features, but the suspensions are different, and two wheelers lean when they turn, which should be easy to distinguish from the accelerometer and gyros.
Seems feasible to detect a user at above foot speed,that appears to be stopping at intersections but seems to be more immune to congestion than others and label it as a two wheeled vehicle.
I did something like this in a quantified self phase, it's fun to see the location patterns. I was gridding locations into buckets and scoring the entropy over time, to see how diverse the set of locations I was going to was. Long story short, not that diverse. Just over 1 bit was enough! (think: home or work)
We actually did that several years ago in our data viz lab. All of the (willing) lab members tracked their location data and one student developed a way to track and visualize co-occurrences of people over time [1].
Another ways to visualize this kind of information is to use space time cube maps, which turn the time axis into the z-axis for trajectory visualization; GeoTime[2] is the most well known implementation of it AFAIK.
Uhmm, maybe it depends on the phone? I backpacked Europe 4 years ago without a data plan and I have all my location history logged, it seems like it kept it on my phone while offline and then it uploaded it when connected, I used a Galaxy S5 btw.
I'd prefer it to be saved locally or sent to a server of mine with some standard protocol (maybe NMEA 0183 over HTTP POSTs) and never to Google. Of course I could write an app for that (and I did some 8 years ago) but it would be much easier if the OS would cooperate.
By the way, I disabled location services on my devices to prevent at least this form of spying on me.
Or you’re getting an IP from somewhere associated with a different region. This happens all the time with my cell connection - it seems the company uses spare Verizon capacity, and when I (or someone else) looks up my 4G is, it’s all over the US with no connection to my actual location.
The probability of each being true independently?
> How can Google possibly predict activity type between IN_TWO_WHEELER_VEHICLE vs IN_FOUR_WHEELER_VEHICLE ?!
Two wheel vehicles can often make it through traffic, in between cars.