3

I'm testing the geocoder results for one state, and I'm getting different results for directions that seem to be well formatted.

E.G.

Geocoding these addresses is taking around 100ms each:

280 FIRST AVENUE,MANHATTAN,NY,1000   
214-17 PALMER DRIVE,QUEENS,NY,11697   
130 LIBERTY STREET,MANHATTAN,NY,10006   
345 PARK AVENUE,MANHATTAN,NY,10154   

While for this group:

53 5TH AVENUE,MANHATTAN,NY,10033  
810 7TH AVENUE,MANHATTAN,NY,10170   
4602 13TH AVENUE,BROOKLYN,NY,1120  

I don't find any clear pattern for to justify this difference, it seems to be happening for all the zip codes:

enter image description here

  • I don't know how TIGER works algorithmically. But do the zip codes have roughly equal numbers of addresses, or are the ones taking longer just really populous (larger n)? – Richard Law Jun 1 '17 at 22:32
1
+50

There is a pattern. The addresses that produce quick results all have ordinal indicators in their street names so it seems that Tiger geocoder has a preference for that sort of format.

Tiger geocoder uses PostgreSQL fuzzy string matching to interpolate and match address along the Tiger edges. PostgreSQL fuzzystrmatch module provides Soundex, Levenshtein and Metaphone functions to determine similarities and distance between strings.

The Soundex and Metaphone systems are methods of matching similar-sounding names by converting them to the same code, therefore you'd be more likely to get extreme matching values with 'th' suffixed street names than you would with generic text and that could in theory lead to quicker results.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.