Label geographic coordinate to district coordinate

[juifa-tsai]

According to pre-EDA, the pickup & dropoff locations can be the significant variables for analysis. Those variables are stored with geographic coordinates, i.e. longitude v.s. latitude. However, those values may be non-meaningful and bias in regression method,, since they are for labeling the location mathematically instead for quantity. Thus, finding the meaningful & efficient way to label the location is an issue. Here is an basic idea - Categorized to belonged districts instead of using continue value :

1. Category types/ranges : Boroughs (5) > Community areas/boards, CB (18max/bor.) > Neighborhoods (?)
2. Label with encode method to binary bits, e.g 010 110 for one of CB (total 59).

Discussion :

1. Should we use multi-categories & single category?
2. Use CB or Neighborhoods?
3. How to extract neighborhoods numbers?
4. How to map geographic coord. to particular district category?

[yennanliu]

Re :
1-3. to - discussion list
4. you can google "polygon ", there are already tools doing this, may be interesting if we can extract features from lon & lat
maybe check :

• http://fiware-orion.readthedocs.io/en/master/user/geolocation/
• https://rstudio-pubs-static.s3.amazonaws.com/257299_79c25a258f9c46e3ab2f3830653195dd.html
• https://stackoverflow.com/questions/31701101/how-do-i-get-region-district-polygon-boundaries-data-from-google-maps

for more information
cheers!

[juifa-tsai]

Or we can define own regions without meaning?
For instance, let's cut the map to many square boxes?

[yennanliu]

yes ! we can, I did similar things before. we can separate the city into group like " most busy",
"busy", "medium", and "casual" ones.

[juifa-tsai]

Cool, I think it quite important do this before diving to ML.
I'd recommend do both:

1. defining box region as chessboard -> for ML
2. label as density -> make chessboard has meaning

What do you think?

I can try to build a function for defining the chessboard & linear distance in this weekend, if you don't mind the time delay lol