Define (subclasses in) ground truth

univie-datamining-team3 / assignment2

Analysis of mobility data

MIT License

0 stars 0 forks source link

Define (subclasses in) ground truth #31

Closed rmitsch closed 6 years ago

rmitsch commented 6 years ago

What's our ground truth? Against which clusters do we want to evaluate? Transport modes is a no-brainer, but after that -

Person collecting the trip data?
Tram/metro line?
Time of the day (could make a difference when taking the tram, not with walking or metro though)?

Do we want want to evaluate against both scripted and unscripted trips?

Lumik7 commented 6 years ago

Another mode most probably would be "idle", which means in this context standing still e.g. while waiting for a traffic light

rmitsch commented 6 years ago

Do you think that being idle/standing still would last long enough to fill up a 30s segment?

Lumik7 commented 6 years ago

For the ordinary trips one could be waiting at a traffic light for one segment

Am 07.01.2018 2:27 nachm. schrieb "Raphael Mitsch" <notifications@github.com

:

Do you think that being idle/standing still would last long enough to fill up a 30s segment?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/univie-datamining-team3/assignment2/issues/31#issuecomment-355822765, or mute the thread https://github.com/notifications/unsubscribe-auth/AOiGfdoOsQKh-cdheWoZPnRmtU3hv4Hnks5tIMYygaJpZM4RPt4K .

rmitsch commented 6 years ago

Alright. So our clusters could be combinations of:

Mode (tram, metro, walk, idle)
Person
Scripted/Unscripted

That'd be up to 24 possible clusters. Automatic evaluation of clustering results might be a little tough, since we don't know how many/which clusters might actually be distinguished. Maybe interactive/visual exploration?

Lumik7 commented 6 years ago

sounds good

Maybe interactive/visual exploration?

what technique would you suggest?

rmitsch commented 6 years ago

I'd probably start with using scatterplots with plotly.