[mkgmap-dev] [PATCH v6] Automatic location completion

Wed Mar 9 22:22:03 GMT 2011

El 08/03/11 23:07, WanMil escribió:
> Next round of the location improvements:
> * The algorithm that searched which elements were contained within a 
> boundary was (is) wrong. I updated some parameters in the Quadtree so 
> I the probability is very much lower that an element is not assigned 
> correctly.
In case they can help you improve the algorithm, here you have two 
examples of nodes wrongly assigned (according to my styles below): nodes 
945168390 and 943005606, that are within boundary relation 348994 
(admin_level=6) are assigned region=Extremadura (boundary relation 
349050, admin_level=4). I can provide more examples if necessary.
Extract of the styles:
mkgmap:region!=* & is_in:province=* { set 
mkgmap:region='${is_in:province}' }
mkgmap:region!=* & mkgmap:admin_level6=* { set 
mkgmap:region='${mkgmap:admin_level6}' }
mkgmap:region!=* & mkgmap:admin_level5=* { set 
mkgmap:region='${mkgmap:admin_level5}' }
mkgmap:region!=* & mkgmap:admin_level4=* { set 
mkgmap:region='${mkgmap:admin_level4}' }
mkgmap:region!=* & mkgmap:admin_level3=* { set 
mkgmap:region='${mkgmap:admin_level3}' }
> * The mkgmap:country, addr:country and is_in:country tag is now always 
> assigned with the three letter country code. This makes it possible to 
> define country specific rules in the style files.
With no changed on your mkgmap:country rules in styles I have (among 
others) Esp and España (ESP) in the list of countries. If I select Esp, 
a full list of places is found, but if I select España (ESP), nothing in 
found.
>
> A note ot Carlos:
> With this patch you can add debug rules to the style file. So if you 
> don't understand why a specific region/country is used add the 
> following line to your style files:
> mkgmap:region=* { echo "R=${mkgmap:region} 3=${mkgmap:admin_level3} 
> 4=3=${mkgmap:admin_level4} 5=${mkgmap:admin_level5} 
> 6=${mkgmap:admin_level6} is_in=${is_in:region}" }
How can I send the echo output to a file? I tried java -jar mkgmap.jar 
options input_files > textfile with no success
>
> I think your rules are ok.
>
> By the way: Should I create a branch for my changes? Maybe this makes 
> it easier for more people to test and to play with?
>
> WanMil
>
>> El 06/03/11 19:40, WanMil escribió:
>>> I implemented Minkos idea using the style file to set the city,
>>> region, zip and country names.
>>>
>>> How does it work?
>>> The new AddressHook adds special tags to each element that is inside a
>>> known boundary:
>>> mkgmap:admin_level2
>>> mkgmap:admin_level3
>>> mkgmap:admin_level4
>>> ..
>>> mkgmap:admin_level11
>>>
>>> If an admin_level is not known the relevant tag is not set. The same
>>> for the zip tag mkgmap:postalcode.
>>>
>>> The style file contains a new rule block (in each file):
>>> mkgmap:country!=* & addr:country=*
>>>    { set mkgmap:country='${addr:country}' }
>>> mkgmap:country!=* & is_in:country=*
>>>    { set mkgmap:country='${is_in:country}' }
>>> mkgmap:country!=* & mkgmap:admin_level2=*
>>>    { set mkgmap:country='${mkgmap:admin_level2}' }
>>>
>>> mkgmap:region!=* & is_in:county=*
>>>    { set mkgmap:region='${is_in:county}' }
>>> mkgmap:region!=* & mkgmap:admin_level6=*
>>>    { set mkgmap:region='${mkgmap:admin_level6}' }
>>> ...
>>>
>>> mkgmap:city!=* & openGeoDB:name=*
>>>    { set mkgmap:city='${openGeoDB:name}' }
>>> mkgmap:city!=* & mkgmap:admin_level8=*
>>>    { set mkgmap:city='${mkgmap:admin_level8}' }
>>> ...
>>>
>>> mkgmap:postal_code!=* & addr:postcode=*
>>>    { set mkgmap:postal_code='${addr:postcode}' }
>>> mkgmap:postal_code!=* & mkgmap:postcode=*
>>>    { set mkgmap:postal_code='${mkgmap:postalcode}' }
>>>
>>> These rules set the mkgmap tags:
>>> mkgmap:country
>>> mkgmap:region
>>> mkgmap:city
>>> mkgmap:postal_code
>>>
>>> Only these tags are used to set the country, the region, the city and
>>> the zip code. No more rules are applied automatically (there are still
>>> some in the Locator class but they will be removed).
>>>
>>> This enables you to configure which admin_level information you want
>>> to use for the city name, the region names etc. Also country specific
>>> rules are possible (e.g. add a mkgmap:country=NLD rule).
>>>
>>> Todo: The country normalization must be performed by the AddressHook
>>> so that one can be sure that the Netherlands are always tagged with
>>> mkgmap:country=NLD (or mkgmap:country=Nederland?)
>>>
>>> The rules are not perfect up to now but please feel free to post some
>>> good rules for your country.
>> I have changed my styles as follows,
>>
>> mkgmap:region!=* & is_in:region=* { set 
>> mkgmap:region='${is_in:region}' }
>> mkgmap:region!=* & mkgmap:admin_level4=* { set
>> mkgmap:region='${mkgmap:admin_level4}' }
>> mkgmap:region!=* & mkgmap:admin_level6=* { set
>> mkgmap:region='${mkgmap:admin_level6}' }
>> mkgmap:region!=* & mkgmap:admin_level5=* { set
>> mkgmap:region='${mkgmap:admin_level5}' }
>> mkgmap:region!=* & mkgmap:admin_level3=* { set
>> mkgmap:region='${mkgmap:admin_level3}' }
>>
>> to adapt them to what I think is best for Spain, but, if I understood
>> well, something is not working as expected: places tagged as villages
>> that have no is_in:region information and are within admin_level=4 and
>> admin_level=6 polygons take the region from the a_l=6 polygon, instead
>> of a_l=4 as it should be from the style above.
>>>
>>> WanMil
>>>
>>>
>>>> Attached is the next step for the automatic location completion:
>>>> * Some speed improvements (it's still slow but not as slow as 
>>>> before ;-)
>>>> * boundary=postal_code is now supported.
>>>> * The Locator does no longer overwrite already set attributes. This
>>>> seems to be a better solution. I think the Locator must be rewritten
>>>> completely. Up to now I have commented some lines only without deep
>>>> understanding what it is really doing.
>>>>
>>>> Some things that don't work:
>>>> * Lot's of POIs are assigned with the default country although the
>>>> MapElements are assigned with the correct country. Example:
>>>> - Spain splitted with --overlap=3000 (more overlap is better for my
>>>> patch)
>>>> - Cafe Antxi (http://www.openstreetmap.org/browse/node/629554219) is
>>>> correctly assigned with country=ESP but in the POI search it is
>>>> displayed with the default country.
>>>> I have no idea why.
>>>>
>>>> * Zips are only applied to POIs. Streets seems to be not yet supported
>>>> by the MDR creation. (=> Steve?)
>>>>
>>>> Things to do:
>>>> * Speed improvement (At the moment the location search takes as much
>>>> time as the rest of the map creation)
>>>> * A better mapping of city, region to the admin-levels
>>>> * Make use more OSM information (like residential areas, place=city 
>>>> POIs
>>>> in residential areas etc.)
>>>> * Clean up of the Locator class
>>>>
>>>> WanMil
>>>>
>>>>
>>>>> I have improved calculation of the country information. In case no
>>>>> country information is available the most used country in a tile is
>>>>> used
>>>>> for all elements without country tag.
>>>>>
>>>>> The patch now patches the trunk and no longer the index branch.
>>>>>
>>>>> I have observed that lots of POIs are assigned the default country
>>>>> although I they are assigned a different country. I have no clue 
>>>>> where
>>>>> this happens.
>>>>> Maybe an index related problem? I was hoping that Steves commit today
>>>>> would fix that problem but it didn't.
>>>>>
>>>>> WanMil
>>>>>
>>>>>> I have done some development and want you to share my current
>>>>>> findings.
>>>>>>
>>>>>> 1. The MapElement copy constructor seems to have a bug. The 
>>>>>> attributes
>>>>>> map which contains the city, region and country information is not
>>>>>> copied. From my point of view this is an important thing that
>>>>>> should be
>>>>>> fixed in the trunk also.
>>>>>>
>>>>>> 2. In the patch the Locator is disabled as much as it was 
>>>>>> possible for
>>>>>> me up to now.
>>>>>>
>>>>>> 3. I am using now separate tags (mkgmap:city, mkgmap:region etc.).
>>>>>>
>>>>>> I recommend to run this patch with location-autofill=-1 or 0 to see
>>>>>> how
>>>>>> the patch works and not how the old Locator works.
>>>>>>
>>>>>> WanMil
>>>>>>
>>>>>>
>>>>>>> After the index branch has been developed so far that we can use 
>>>>>>> the
>>>>>>> results in MapSource (thanks to Steve!!) there is a big need for a
>>>>>>> mechanism to fill missing location information.
>>>>>>>
>>>>>>> There has been a lot of discussion about that. It stopped 
>>>>>>> because it
>>>>>>> was
>>>>>>> quite theoretical (from my point of view). Therefore I have
>>>>>>> started to
>>>>>>> implement a prototype osm hook that tries to complete the missing
>>>>>>> location information. I think the is_in tag is quite problematic
>>>>>>> therefore I decided to use the addr tags.
>>>>>>>
>>>>>>> Attached is a *very* early implementation that makes use of the
>>>>>>> boundary
>>>>>>> polygons. It checks all nodes and ways/polygons within all boundary
>>>>>>> polygons and fills the missing addr tags.
>>>>>>>
>>>>>>> I think this is a better foundation for the discussion how we 
>>>>>>> want to
>>>>>>> fill the missing location information.
>>>>>>>
>>>>>>> Before flaming me about terrible code please be aware that:
>>>>>>> * the patch is not optimized
>>>>>>> * it's version 1, so far away from handling all situations /
>>>>>>> countries
>>>>>>> * it's a very early prototype
>>>>>>>
>>>>>>> I would be happy if you try it and make some proposals how it
>>>>>>> could be
>>>>>>> improved.
>>>>>>>
>>>>>>> Some improvements I already know and thinking about:
>>>>>>> * The new hook should run after the multipolygon finishing and not
>>>>>>> before
>>>>>>> * The is_in tag should be handled
>>>>>>> * boundary information propagation => First go through all 
>>>>>>> boundaries
>>>>>>> and add the information from higher levels to lower ones, i.e. add
>>>>>>> addr:district to all city boundaries within that district
>>>>>>> * Use landuse=residential information in combination with the place
>>>>>>> nodes
>>>>>>> * In case a boundary level is missing (which mostly will be the 
>>>>>>> case
>>>>>>> for
>>>>>>> countries admin_level=2), select the addr information from all 
>>>>>>> nodes
>>>>>>> and
>>>>>>> ways within a boundary by using the most used addr-tag value. (in
>>>>>>> germany the addr:country=DE should be the most used)
>>>>>>>
>>>>>>>
>>>>>>> Have fun!
>>>>>>> WanMil
>>>>>>>
>>>
>>>
>
> _______________________________________________
> mkgmap-dev mailing list
> mkgmap-dev at lists.mkgmap.org.uk
> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev


-- 
Por favor, no me envíe documentos con extensiones .doc, .docx, .xls, .xlsx, .ppt, .pptx, .mdb, mdbx
Instale OpenOffice desde http://es.openoffice.org/programa/index.html
OpenOffice es libre: se puede copiar, modificar y redistribuir libremente. Gratis y totalmente legal.
OpenOffice está en continuo desarrollo y no tendrá que pagar por las nuevas versiones.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mkgmap.org.uk/pipermail/mkgmap-dev/attachments/20110309/134727b7/attachment.html