search

datamapio / posmo_one_testing

Public Testing Repo
0 stars 0 forks source link

Wrong routing #33

Open thohug opened 3 years ago

thohug commented 3 years ago

Setting Chur, Switzerland, Android

wifi-fence: 1 do_not_track: 0 wifi_over_gps: 1 data_over_wifi_only: 1

Issue Originally this part (2021-1-1) was processed as a car segment (I have already changed it to train). The car was more or less routed from train stop to train stop - resulting in these weird loops -, even though the data gathered from the train was quite good the routing was way off...

grafik

grafik

rogerfischer commented 3 years ago

This is the map matching that does these crazy things. I wouldn't use it for now, @djr4 @aelvyr, as it creates more problems than it solves. Also if someone reclassifies, we need to rerun the processing for that part, otherwise we stick with street matching for trains as in this example.

PS: @thohug As soon as we have the GTFS-API ready this should be solved.

thohug commented 3 years ago

I feel like map matching could be quite useful, but in this case it's probably done more harm as it "overwrote" the (good) train data as it matched it to car... Probably there needs to be some good weighting between mapmatching and real data.

rogerfischer commented 3 years ago

Yes, we have to rerun your day without map matching.

aelvyr commented 3 years ago
Screen Shot 2021-01-04 at 4 50 08 PM

If we reprocess that segment, it is actually now recognized as a train, and there is no map-matching issue. There are some newer changes that have not yet been deployed that make this possible. About new deployments: we will only able to make new deployments from next Monday on.

Let me know if I should reprocess the day to your account and UI.

thohug commented 3 years ago

Looking good to me. Maybe a refresh button would be a good idea to reprocess a day once the user made some changes?

aelvyr commented 3 years ago

Ok, so I will reprocess this for you. And yes agreed something like that is necessary, but there are also some difficulties that come with it.

thohug commented 3 years ago

great - btw it's looking like I'm having a similar one today (wrong bus map matching instead of train).

aelvyr commented 3 years ago

I see, yes we have a currently updated model for transport mode prediction, which is much more reliable, so once we redeploy a new backend, wrong transport modes should be improved, but there might still be some things that can't be recognized perfectly without a transport mode context (like station detection, which we are currently working towards). So feel free to write an issue about anything that is not possible to relabel and validate until we have a possibility for users to reprocess themselves.

aelvyr commented 3 years ago

Also let me know if there is anything wrong with the reprocessed data

thohug commented 3 years ago

Another example of map matching pretty much ignoring the actual route (data by bike) grafik