meeting-agendas/agendas/20220608Agenda

[utterances-bot]

DaanMatch - Check-in — Meeting Agendas

https://daanmatch.github.io/meeting-agendas/agendas/20220608Agenda.html

[shpatrickguo]

Geocoding Addresses

Reached out to Apoorv for this project.
Complete

You can find the Geocode repo with the script on Geocode API.
As I pointed out last year, using open source APIs will lead to a lot of failures. In fact, as of right now, out of 170,000 addresses I have yet to come across a single success. D:
Nevertheless, it is not feasible for the organization to spend money on the Google API/paid alternatives.

I came up with a couple solutions:

1. Since the open source APIs cannot deal with the addresses in such granularity. We will truncate the addresses to capture the locality.
   For example:

# Failure will original address
original_addr = "Room A-528, Near Dayal Market, Narela Road, Alipur, New Delhi, Delhi- 110036"

# Success with truncated address
new_addr = "Alipur, New Delhi, Delhi- 110036" 
OUTPUT = ('28.7959955', '77.1360706')

2. Use a series of open source APIs, and create a parent function to try out all of them

def geocode(addr):
    """ Geocode address to latitude and longitude using multiple apis and return non-failure.
    
    INPUT: string - address
    OUTPUT: tuple - (latitude, longitude), or (Failed, Failed)
    """
    openstreetmaps(addr)
    geopy(addr)
    API3(addr)
    
    return Non-failure output, if all fail, return failure

This will allow us to explore:

• Which API performs the best
• Percentage of failures
• Any patterns in the addresses of the failures e.g. the address row is too short or has weird characters, no pincode etc.

If this is complete

1. Will pass along additional lat long to Jarrett.
2. Update database/datastore with lat long information

Caveats

Will not be able to use previous lat long values, because we are using different sources. So everything from the Google API cannot be used.