Respondida

Auto flag cyclists in run segments

Hi, would it be possible to automatically flag people who cycle on run segments. Anyone moving at world record mile pace or above usually gives it away

51

Comentarios

40 comentarios
  • Comentario oficial

    We already have an auto-flag system in place. This works based on whether the uploaded activity ranks first on any segment leaderboards above a certain % threshold which is considered unlikely. It's possible that the current system misses auto-flagging some activities that are not runs, and any examples would help us determine improvements we could make. Thanks! 

    Acciones de comentarios Permalink
  • It's incredibly easy to set data validation to make this happen. There is no reason why this shouldn't be implemented already.

    13
    Acciones de comentarios Permalink
  • Totally agree buddy!

    2
    Acciones de comentarios Permalink
  • I too think that it would be a good idea to incorperate this into Strava. I have noticed this happen on many occasions. It seem to be an honest mistake made by new users, if they receive an automated response that the activity has been changed to the correct one it would help things run much smoother.

    4
    Acciones de comentarios Permalink
  • Agreed

    3
    Acciones de comentarios Permalink
  • This would also be good for cycling when some uphill or long flat segments could only be achieved by a car or motorbike

    4
    Acciones de comentarios Permalink
  • The smartphone app ought to be able to distinguish between running and cycling easily enough using the accelerometer. Running bounces up and down in a regular cadence whereas cycling has a smoother baseline with irregular bumps.

    The other thing that happens a lot is someone runs a few miles and then gets in the car to drive home without stopping Strava. It should be easy to filter out any segments after a certain threshold speed has been reached.

    3
    Acciones de comentarios Permalink
  • Yea, it's crazy to me that this isn't implemented yet. I have several (very difficult) running segments on mixed-use paths, and it seems like every day I get emails telling me I've lost my CR. 

    Strava - if someone sets a national record anywhere during the run, flag it. Even if it's wrong 1% of the time - when you think something's sketchy, send an email to the athlete that says "Hi - looks like something may have been wrong with your run, could you check it?". 

    5
    Acciones de comentarios Permalink
  • Please fix this!

    4
    Acciones de comentarios Permalink
  • Totally agree

    2
    Acciones de comentarios Permalink
  • Just cancelled Strava Premium, as this is the most important feature that looks like it will never be introduced.
    I did a run yesterday and checked my progress on a few segments only to find 80+ cyclists registering impossible "running" times on the segments.

    I am not going flagging 80+ activities, Strava should be auto flagging better than world record running activities.

    11
    Acciones de comentarios Permalink
  • Why no updates on this? It would be such a simple thing to fix.

    2
    Acciones de comentarios Permalink
  • This really needs to happen. I can understand there being some issues where a slow cyclist could be confused for a fast runner but as others have noted it it's pretty common to have a large number of folks 'running' 1:30 min/mi pace through segments.  Or even running a high 3 min/mi pace but it's for 20+ miles.  It should be easy to at least auto-flag these since presumably someone isn't running 2-3 marathons in a row at a sub 4 minute pace.

    You have made the flagging system easy to use, for which I thank you,  but it's such a prevalent problem that if is super time consuming to police segments.  I would really love to see segments become as popular in running as they are in cycling but things like this need ironed out for it to happen.

    2
    Acciones de comentarios Permalink
  • Elle - Can you see the history of "runs" I've flagged? It's >50 per year, and they're always because it's a cyclist that has taken a segment. That should be a decent data set to start with... if not, here are some that I got recently:

     

    https://www.strava.com/activities/1092034314

     

    https://www.strava.com/activities/1102560068/overview

     

    https://www.strava.com/activities/1107696262/overview

    3
    Acciones de comentarios Permalink
  • Hello Elle, where I have noticed it happening quite frequently is when new segments are created.  I don't have any examples now because i've cleaned up all of our local ones.

     

    Here is a segment with an obvious cyclist taking the CR, though it was from 2 years ago and I don't know when the segment was created: https://www.strava.com/segments/11958037

    1
    Acciones de comentarios Permalink
  • Elle, here's one from today:

     

    https://www.strava.com/activities/1157258235/segments/28508205434

     

    He took three segments with that "run", including a 0.9 mile segment at 2:41 pace. If auto flagging exists, I can't imagine why that'd make it through.

    2
    Acciones de comentarios Permalink
  • @ Todd, that's because there is no auto flag feature, or if there is, its badly broken, it can't tell the difference between a "run" done in a vehicle and a crawl

    1
    Acciones de comentarios Permalink
  • Elle - Here's another one:

     

    https://www.strava.com/activities/1162695522/segments

     

    He took three segments on that ride, all of them well under 4 minute pace (and the one he took from me was 2:35 pace!, more than 40% faster than the previous CR).

    3
    Acciones de comentarios Permalink
  • The auto flagging system is broken. For example, this activity holds at least one segment CR. But the mile split times are unrealistically fast and the overall pace is near the marathon world record even with all the hills. Obviously he was cycling or using some kind of vehicle.

     

    https://www.strava.com/activities/1124082116/overview 

    0
    Acciones de comentarios Permalink
  • Here's another one. Strava didn't auto flag a 1:46 mile run pace? Seriously? This is pathetic.

     

    https://www.strava.com/activities/1122637227/overview 

    1
    Acciones de comentarios Permalink
  • I agree - each time I look through the leaderboards for the challenges there are people there faster than world record pace for 5k and HM etc.  So I don't think the autoflag is working.at all

    5
    Acciones de comentarios Permalink
  • 0
    Acciones de comentarios Permalink
  • Here's another one.

    https://www.strava.com/activities/976155624/overview

    A 2:40/mi running pace? How could Strava fail to auto flag something ridiculous like that? Is anyone at Strava actually paying attention to this stuff?

    0
    Acciones de comentarios Permalink
  • An Auto Flag Function would never be able to be perfect, but just implementing some basic logic should be able to take care of the most common/blatant cases.  It will always be difficult to auto distinguish someone leisurely riding a bike on a path as a bike or run activity, but the blatant stuff should be easy to pick out.

    - Anything classified as a run should be flagged if: at any point max speed exceeds 25 mph for more than 100 meters, if any World Records are broken...(specifically for a Mile, 10k, or Marathon to start with). 

    Or, based on the number of historical flags a user has, mark them as likely offenders that should have less leniency on the sensativitly of the auto flag checks.

     

    1
    Acciones de comentarios Permalink
  • Here's another one.

    https://www.strava.com/activities/1250864151/overview

    Obviously no one can run a 2:35/mi pace for 3 miles.

    Is anyone at Strava paying attention?

    0
    Acciones de comentarios Permalink
  • 1
    Acciones de comentarios Permalink
  • > This works based on whether the uploaded activity ranks first on any segment leaderboards above a certain % threshold which is considered unlikely. 

    That is an incredibly unreasonable way to do things since it should be done on absolute speed and not a percent of the threshold. For example, this activity shows the "runner" reaching about 40mph. By not auto-flagging events you're creating hundreds of hours of work for users per week when a single engineer at Strava could place a threshold in under two hours to eliminate that. It shows that your developers don't value the time of your users.

    https://www.strava.com/activities/1390100355

    About about these distances and times to be auto-flagged. The false positive rate will be once every few year and can be manually reviewed.

    800M+ - Over 17mph

    Mile+ - Over 15mph

    Half marathon+ - Over 14mph

     

    2
    Acciones de comentarios Permalink
  • > This works based on whether the uploaded activity ranks first on any segment leaderboards above a certain % threshold which is considered unlikely.

    The threshold sounds to me like a reasonable approach, because it's more robust to segment peculiarities, e.g., created with bad data, screamingly downhill, passing through a block of skyscrapers. However, it doesn't seem to get triggered very easily, so maybe it's set too low? If I break a kom by 10%+, I would find it amusing, rather than frustrating, to have to explicitly confirm the activity was a run. But, anyway, that is only likely to happen on segments with a couple dozen attempts. So, perhaps the threshold should adapt to the number of attempts, such that mature, popular segments auto-flag record-breaking performances more eagerly?

     

    That said, an approach that compares to the kom based on a threshold seems inherently limited, because it's only going to catch those that are easy to spot/verify manually, anyway. The more time consuming ones are the "3:00/km on an uphill segment" or "2:30/km on a downhill segment" type efforts, which are plausible but exceedingly unlikely. Often, one cannot even verify whether that's a legitimate attempt by an unknown superstar, because the profile is private so the activity cannot be contrasted to the athlete's normal performance range. Yet, this is where Strava can automate the process by comparing to the athlete's previous attempts on a segment. If the threshold is applied not vs just the kom but also vs the athlete's best effort on that segment, that will have three advantages:

    1. 1. Somebody flagged for one ride/ski/rollerski on a segment will get auto-flagged easier on subsequent misclassified ones, as their best run performance on that segment will slow down
    2. 2. It will apply throughout the leaderboard, not just to the kom
    3. 3. It will catch some of those borderline cases where it could plausibly be a legit run by an unknown star so can't be confidently flagged by other users and therefore softens the vs-kom-threshold test.

     

    1
    Acciones de comentarios Permalink
  • +1 for more robust auto-flagging. It's annoying to have to flag stuff yourself every day when it's really, really obvious what's going on. I get wanting to avoid false positives, but we're a long ways from that right now. Also, it seems like the consequences of a false positive are small--the user gets a chance to say, "yes, this is in fact real."

    3
    Acciones de comentarios Permalink
  • Agreed!

    Haven't seen that Strava auto-flag system yet...

    3
    Acciones de comentarios Permalink

Iniciar sesión para dejar un comentario.

¿No encontró lo que buscaba?

Nueva publicación