Discussion: GC Websync Activity Type
in: Orienteering; The Website
May 30, 2018 8:16 PM
#
How does AP learn what activity types you do and automatically assign them? The fact that it knows I've been cycling is quite obvious. But the fact it knows I've been orienteering still amazes me. Tonight I went for a road run, which I've only just started logging separately to non-tarmac runs (just so I can keep track). I've only done this a few times and it's already learned!
May 31, 2018 1:37 AM
#
Perhaps it's the slow-downs to punch? Or laps if you take them while orienteering?
May 31, 2018 2:47 AM
#
probably syncing the location and time with a known calendar of events
May 31, 2018 3:07 AM
#
An army of gnomes in Ken's basement that follow each Attackpoint user around on their workouts then assign the activities when they're uploaded
May 31, 2018 3:21 AM
#
Nothing to do with the calendar. I went for a warm up run to an unofficial event site yesterday then ran that and it knew to categorise one as a run and the other as urban orienteering. It possibly recognises patterns in movement as well as different speeds compared to your past performance. The 'orienteering' component of the run didn't have any slow downs (nothing physical to punch) although I did record splits.
My MTB ride on Tuesday though was classified as MTB orienteering although it's a route I've done many times before and always put down as MTB so it's not foolproof!
My MTB ride on Tuesday though was classified as MTB orienteering although it's a route I've done many times before and always put down as MTB so it's not foolproof!
May 31, 2018 4:00 AM
#
ken:
The general idea is that AP computes a number of feature characteristics about each of your sessions and activity types. It can then use the distributions of those values to classify new sessions as whichever of your activity types they are most likely to be (with mixed success).
The actual features are all related to things in the data file like speed, grade, heart rate, cadence, power output, and a few other contrived values like a measure of how straight or twisty the track is. It doesn't consider the location at all, except to compute those secondary things about the shape of the track.
Originally I tried gnomes, but they were moody and too expensive to feed.
The actual features are all related to things in the data file like speed, grade, heart rate, cadence, power output, and a few other contrived values like a measure of how straight or twisty the track is. It doesn't consider the location at all, except to compute those secondary things about the shape of the track.
Originally I tried gnomes, but they were moody and too expensive to feed.
Jun 1, 2018 11:55 AM
#
I've been impressed by this for ages as it's so accurate. I kinda guessed it used things like overall speed, speed variation but would never have guessed twistyness of the track!
Awesome modelling Ken!
Awesome modelling Ken!
Please login to add a message.