2 ms·

We tried to implement mapbox on a new product last week. I like the style options for the maps but their reverse geo lookup and local search performed poorly vs

We tried to implement mapbox on a new product last week. I like the style options for the maps but their reverse geo lookup and local search performed poorly vs google maps. Pins dropped down the street from actual address and search results don't narrow in accurately using POI and city/state - for example "Chipotle Parker Colorado" rather than showing the 3 chipotles in Parker, CO lists: - the city "Parker, CO" - Parker street in Pueblo, CO - childrens hospital therapy center in Parker, CO It just doesn't match the behavior our users have - probably because they have been trained how to search using google and google maps. I am guessing the main use for mapbox is actually custom tiles and custom datasets but anyone using it as a replacement for google maps I'd love to hear if you were able to solve this without too much trouble. We preferred their API, sdks, and pricing over google maps but just couldn't get the results we wanted. https://docs.mapbox.com/search-playground/ https://docs.mapbox.com/search-playground/

lukeqsee 6y ago
Geocoding is really hard. :) https://opencagedata.com https://opencagedata.com seems to be the best around using open data and offer API compatibility (or did at last check).
et-al 6y ago
afaik, Mapbox's geocoding service was never meant to provide the lng,lat of retail entities. Looking over their API docs, you can search for points of interest (POI), but that's it: https://docs.mapbox.com/api/search/geocoding/ https://docs.mapbox.com/api/search/geocoding/ For your use case, you should look into a geocoding service [0]. And then you can display the location results with your mapping library of choice. Google Maps does provide an all-inclusive service, but as you've mentioned, styling is better with Mapbox (and pricing). [0] - Aside from Google and Mapbox, I've never used any of these alternative geocoding services: https://smartystreets.com/articles/is-there-a-geocoding-api-as-good-as-google https://smartystreets.com/articles/is-there-a-geocoding-api-...
- harrisonjackson 6y ago
  > The Mapbox Geocoding API does two things: forward geocoding and reverse geocoding. Forward geocoding converts location text into geographic coordinates, turning 2 Lincoln Memorial Circle NW into -77.050,38.889. https://www.mapbox.com/place-search https://www.mapbox.com/place-search I think "location text" is the part they are not parsing well. The screenshots all over that page show someone entering a couple of letters and getting the resulting POI with address. Unfortunately, we could not get reliable results using their web place search so maybe they just have better mobile sdks that are adding more filter data to the requests automatically ¯\_(ツ)_/¯
  - structural 6y ago
    Their geocoding seems to work decently enough for street addresses from what I've seen, which is what it's technically advertised to do. It doesn't have a Google-sized database of business entities, so I'm not surprised it doesn't work for any POI less significant than a major landmark (i.e. "city hall", "the statue of liberty", etc.) Pretty common to use Google to look up an entity (to either get address or full geocoding results) and then Mapbox to do the display though.
  - et-al 6y ago
    You're right, Mapbox does claim to search for businesses. If you're sticking with Mapbox, you'd probably need to a two-step lookup to obtain some useful results. 1. If there's no street address, check the search string to see if there's a city, or city, state. Find the lng,lat of that to use as the proximity bias. If nothing exists, use the user's current location. (Maybe this is why the mobile SDK is better.) 2. Then add the proximity bias while searching for the POI But just a heads up, even a simple city query can be problematic. From personal experience, "san jose" would return San Jose in Costa Rica, or San Jose City in the Philippines, and rarely San Jose, CA, US unless a user was to type all that. Google definitely has better NLP to handle user search queries.
juliansimioni 6y ago
Try us out at Geocode Earth! (https://geocode.earth https://geocode.earth) We wrote and maintain the open-source Pelias geocoder(https://github.com/pelias/pelias https://github.com/pelias/pelias) at Mapzen and have been continuing to improve it in the 3 years since Mapzen's shutdown. Geocode Earth is our hosted service built around Pelias so you can get started quickly, not worry about keeping data up to date, or doing all the fine tuning and optimization that we've had to do to get fast and reasonably accurate autocomplete outside of Google and Mapbox. Using our hosted comparison tool you can see that we do a pretty decent job of finding two out of 3 of the Chipotles (the third might have been deduplicated or not in OSM): https://pelias.github.io/compare/#/v1/autocomplete?text=Chipotle+Parker+Colorado https://pelias.github.io/compare/#/v1/autocomplete?text=Chip... Most our users end up passing a focus point, in which case the results are generally even better, although they're the same in this example https://pelias.github.io/compare/#/v1/autocomplete?focus.point.lat=39.51614762331245&focus.point.lon=-104.76572744432639&text=Chipotle+Parker+Colorado&debug=1 https://pelias.github.io/compare/#/v1/autocomplete?focus.poi...
- harrisonjackson 6y ago
  Nice - tested a bunch of the ones we were benchmarking against and they all worked well! We will keep you in mind as our usage grows or if we have another project. Your tiered pricing doesn't make sense for us as we launch.
  - juliansimioni 6y ago
    Great, glad they worked! Thanks for the good feedback on our pricing. We plan to roll out pay-as-you-go pricing in 2021 that should help with that. In the meantime feel free to reach out to us at hello@geocode.earth and we can set you up with a smaller plan to get you started, if you'd like.

lukeqsee 6y ago

Geocoding is really hard. :) https://opencagedata.com https://opencagedata.com seems to be the best around using open data and offer API compatibility (or did at last check).

et-al 6y ago

afaik, Mapbox's geocoding service was never meant to provide the lng,lat of retail entities. Looking over their API docs, you can search for points of interest (POI), but that's it: https://docs.mapbox.com/api/search/geocoding/ https://docs.mapbox.com/api/search/geocoding/ For your use case, you should look into a geocoding service [0]. And then you can display the location results with your mapping library of choice. Google Maps does provide an all-inclusive service, but as you've mentioned, styling is better with Mapbox (and pricing). [0] - Aside from Google and Mapbox, I've never used any of these alternative geocoding services: https://smartystreets.com/articles/is-there-a-geocoding-api-as-good-as-google https://smartystreets.com/articles/is-there-a-geocoding-api-...

harrisonjackson 6y ago
> The Mapbox Geocoding API does two things: forward geocoding and reverse geocoding. Forward geocoding converts location text into geographic coordinates, turning 2 Lincoln Memorial Circle NW into -77.050,38.889. https://www.mapbox.com/place-search https://www.mapbox.com/place-search I think "location text" is the part they are not parsing well. The screenshots all over that page show someone entering a couple of letters and getting the resulting POI with address. Unfortunately, we could not get reliable results using their web place search so maybe they just have better mobile sdks that are adding more filter data to the requests automatically ¯\_(ツ)_/¯
- structural 6y ago
  Their geocoding seems to work decently enough for street addresses from what I've seen, which is what it's technically advertised to do. It doesn't have a Google-sized database of business entities, so I'm not surprised it doesn't work for any POI less significant than a major landmark (i.e. "city hall", "the statue of liberty", etc.) Pretty common to use Google to look up an entity (to either get address or full geocoding results) and then Mapbox to do the display though.
- et-al 6y ago
  You're right, Mapbox does claim to search for businesses. If you're sticking with Mapbox, you'd probably need to a two-step lookup to obtain some useful results. 1. If there's no street address, check the search string to see if there's a city, or city, state. Find the lng,lat of that to use as the proximity bias. If nothing exists, use the user's current location. (Maybe this is why the mobile SDK is better.) 2. Then add the proximity bias while searching for the POI But just a heads up, even a simple city query can be problematic. From personal experience, "san jose" would return San Jose in Costa Rica, or San Jose City in the Philippines, and rarely San Jose, CA, US unless a user was to type all that. Google definitely has better NLP to handle user search queries.

juliansimioni 6y ago

Try us out at Geocode Earth! (https://geocode.earth https://geocode.earth) We wrote and maintain the open-source Pelias geocoder(https://github.com/pelias/pelias https://github.com/pelias/pelias) at Mapzen and have been continuing to improve it in the 3 years since Mapzen's shutdown. Geocode Earth is our hosted service built around Pelias so you can get started quickly, not worry about keeping data up to date, or doing all the fine tuning and optimization that we've had to do to get fast and reasonably accurate autocomplete outside of Google and Mapbox. Using our hosted comparison tool you can see that we do a pretty decent job of finding two out of 3 of the Chipotles (the third might have been deduplicated or not in OSM): https://pelias.github.io/compare/#/v1/autocomplete?text=Chipotle+Parker+Colorado https://pelias.github.io/compare/#/v1/autocomplete?text=Chip... Most our users end up passing a focus point, in which case the results are generally even better, although they're the same in this example https://pelias.github.io/compare/#/v1/autocomplete?focus.point.lat=39.51614762331245&focus.point.lon=-104.76572744432639&text=Chipotle+Parker+Colorado&debug=1 https://pelias.github.io/compare/#/v1/autocomplete?focus.poi...

harrisonjackson 6y ago
Nice - tested a bunch of the ones we were benchmarking against and they all worked well! We will keep you in mind as our usage grows or if we have another project. Your tiered pricing doesn't make sense for us as we launch.
- juliansimioni 6y ago
  Great, glad they worked! Thanks for the good feedback on our pricing. We plan to roll out pay-as-you-go pricing in 2021 that should help with that. In the meantime feel free to reach out to us at hello@geocode.earth and we can set you up with a smaller plan to get you started, if you'd like.