search

bcgov / ols-geocoder

Physical Address Geocoder
Apache License 2.0
10 stars 6 forks source link

Investigate addresses with lower match accuracy than 4.0 #160

Open bstratto opened 3 years ago

bstratto commented 3 years ago

Some addresses are getting lower match accuracy in 4.1 testing. This issue is to provide examples of such addresses. Examples will be added as comments.

bstratto commented 3 years ago

Address: 8924 ROCKY PINES IR LOWER NICOLA, BC Release 4.1 standardizes to 8924 Rocky Creek Rd, Ladysmith, BC. LOWER NICOLA is a known locality. ROCKY PINES is an IR in the LOWER NICOLA

cmhodgson commented 3 years ago

I get the same result and score for 8924 ROCKY PINES IR LOWER NICOLA, BC in both 4.0.1 on prod as on 4.1.0 on dev.

mraross commented 3 years ago

Me too. The results are the same in both versions. We couldn't make 4.1 better because we couldn't figure out how to make the frontGate optional. Even if we could, I couldn't find a Rocky Pines IR in Lower Nicola. I did find the following IR:

Nicola Mameet 1 -- Lower Nicola, BC

Carla Jack, the provincial toponymist, did say reserve names have been changing and the feds are behind in giving her updates so maybe that's why we don't have rocky pines ir yet.

@bstratto How did you verify that Rocky Pines IR -- Lower Nicola exists?

bstratto commented 3 years ago

Hi @mraross, I googled rocky pines and saw they have a Community Centre. This was not in an official government site though, was in a local news site: https://www.q101.ca/news/12483-community-centre-to-benefit-all-that-live-at-rocky-pines This slide deck shows Rocky Pines as part of the Lower Nicola IRs, but it looks like it is a neighbourhood, not an IR on its own right. See slide 18. Sorry for the confusion. I think this is a non-example.

bstratto commented 3 years ago

This I believe is a pertinent example: Address: 3938 INDIAN RIVER DR VAN, BC 4.1: Van Anda, BC 4.0: 3938 Indian River Dr, District of North Vancouver, BC

mraross commented 3 years ago

Very good example. Very weird. When I enter your example into the 4.1 demo app, the top-ranked autocompleted suggestion is:

 3938 Indian River Dr, District of North Vancouver, BC

but when I hit the search icon, it gives me:

 Van Anda, BC
mraross commented 3 years ago

The geocoder will only do one locality hop so if there is an alias called "Van" that maps to Vancouver, the geocoder won't look further than Vancouver. If we define more aliases for Van (e.g., District of North Vancouver), then it should find the correct address.

bstratto commented 3 years ago

Another case: Address: 35378 IRON ROAD PRINCE GEORGE, BC 4.1: 35378 Iron Rd, Woodpecker, BC , score:63 4.0: 35378 Iron Rd, Woodpecker, BC , score 65 Faults reported by 4.1: "PRINCE GEORGE" LOCALITY.spelledWrong 2 "PR GEORGIA" LOCALITY.notMatched 35 Based on proximity on the map, this may be a locality alias candidate: Prince George as alias for Woodpecker

mraross commented 3 years ago

locality alias:

  Prince George => Woodpecker

has been added to geocoder 4.1 silver at Refractions. Now the following address:

 35378 Iron Rd, Prince George, BC

returns:

 35378 Iron Rd, Woodpecker, BC

with a score of 96.

bstratto commented 3 years ago

Another example: Address: 1832 UP AMBROSI RD, KELOWNA, BC 4.1: Kelowna, BC 4.0: 1832 Ambrosi Rd, Kelowna, BC Faults reported by 4.1: "UP AMBROSI" | STREET_NAME.notMatched | 12

alixcote commented 1 year ago

Tagged with milestone 4.3 to review test cases and if any improvements.

BK01 commented 11 months ago

Results in Geocoder 4.3 are consistent with the results from Geocoder 4.1 as listed above. We will also run the full Address List Match Accuracy test on Geocoder 4.3 and post the resulting score summary in https://github.com/bcgov/ols-geocoder/blob/gh-pages/address-list-match-accuracy.md