README_todo.md

TODO

Sythetic data

1. See it's own README.md
2. Specifically, create trajectory class to store all the data
   • Use the functions from the scripts and have a few get/set methods
3. Write method to output CSVs and work from pandas DFs

Illustration

1. $\checkmark$ Find the correct NBs
2. $\checkmark$ Grab the figures
   • with the overall problem
   • with the box method solution
3. $\checkmark$ Include realworld example with map and 2D1T plot
   • $\checkmark$ make the connection to the 1D1T plots
4. Revise and correct the text
5. $\checkmark$ Create a final html NB $\to$ moved to the root README.md

Evaluation methods

1. Turn evalutation metrics into module Note see the quick_box NBs

Box methods

1-way box methods

basic box method (currently as quick_box; initially as bounding_box)

1. Update and clean up
   • there may be some redudancy in the scripts $\to$ check this!
2. apply batch evaluation
   • Note this method doesn't distinguish travels so well as it looks for mini-clusters
     • it has a generically high false-pos. rate.

metric box method (currently as cluster_splitting)

1. Revisit and clean up
2. check if it is more useful than the basic box
   • ideally if it gives a better starting point than the basic box method
3. Apply the evaluation
   • is it better than the above method
4. Document it's limitations

2-way box method (currently as box_method)

1. Refine algorithm and make it robust
2. Ensure that it produces what it should
   • Test it with the evaluations
3. Include previous box methods as initial passes to refine upon
   • this will give a speed-up but also improve accuracy.
   • will require merging and checking
   • Some ideas:
     1. first pass with quick_box
     2. apply 2-way box to longest cluster
        • check if the box increases
        • check slope
     3. iterate over all remaining clusters
        • start with next longest
        • skip if they overlap with a newly processed cluster

Other methods

1. Piecewise linear

   • can it be made to converge (and faster)?
   • can it be adopted into the box method(s) to improve speed?
     • should it?
   • can it hold up in 2D?

2. DB-Scan

   • rename as "DB_split"
     • since the DBSCAN clusters will be split by
       • the time requirement
       • inbetween cluster(s)
   • try again with the random data in batch
   • apply the evaluation