2012/12/23 Glenn Plas <gl...@byte-consult.be> > Sander/Jo > > Lovely stats, but I don't really agree with all the conclusions, plenty of > doubtful assumptions as well stated, but I will go along with those for a > second. > > I work professionally with OSM data, more specifically Nominatim for years > now(2007/8), If you do not account for the house numbers, the results are > far better than google(is)/yahoo(was). > > The engines that perform the best reverse geocoding results in order : > nominatim, google, yahoo, bing, ...rest > > In reality the only thing that is preventing a full geocode (reverse) > result is the missing housenumber. The streets geocode fine, with better > quality than the number 1 (google at this moment). Professionally I don't > care too much for house numbers (albeit I understand the need for this in > other applications of course). All commercial ones have easter eggs > everywhere, I found more than 20 this year in google, streets that don't > exist so they can see if they get copied over. These will be returned in > geocode results. > > So forgive me, but reading that I will only get 1 hit out of 40 is just > not correct compared to reality. I am able to geocode everything I throw > at nominatim, I'm not talking about 1000 geocodes a week but millions (5 > to 10). Only 5 % is being relayed to a different engine since the DB I > talk about is a Benelux OSM dump, not a full, so anything outside the > benelux will be sent to external engines. > > I would accept those numbers if you would state: "Getting back a full > address including housenumbers will probably work 1 out of 40" But you WILL > get an address back from Nominatim in all other cases, without a house > number of course. > > It surprises me a lot that I'm even in those top stats with a mere 111 > addresses added.... I don't think we can talk about a success in .BE > looking at those totals. But I do agree its improving, compared to 2008 > nominatim result sets. it not only improved in source quality (OSM) but > also implementation (so postal code is returned now, it didn't use to do > that). > > Happy holidays! > > PS: I'm using a lot of geocoding when I truly mean 'reverse geocode', I > only focus in getting full addresses back from coordinates. >
That's probably the main difference. I focus on geocoding (getting coordinates from an address). And when you have long streets without housenumbers, Nominatim will give a result, but the result can easily be 20 km off. Which is really unacceptable. As you don't know how long the street is, you can only assume that you have the correct place when the housenumber is found. So that's what I meant with the "1 in 40" statement. And I agree that the data is becoming better and better. When I started, my own street wasn't on the map (this triggered me). Now the street network is good as complete (we even have streets in the DB that are still under construction). So we need to focus on other things. Those things are (according to me): - Addresses including housenumbers - POI's including extra information s.a. opening hours - Extra info on roads (maxspeed, lanes ...) I just wanted to see how our addresses are evolving, as a result of the other blog posts.and it wasn't that hard to get some indicative numbers. Regards, Sander
_______________________________________________ Talk-be mailing list Talk-be@openstreetmap.org http://lists.openstreetmap.org/listinfo/talk-be
