Commit Graph

82 Commits

Author SHA1 Message Date
Al
ca25b48687 [fix] Not writing empty fields in formatted addresses 2015-09-22 08:13:55 -04:00
Al
134cf616d6 [osm] Using street for language disambiguation in training data 2015-09-21 04:09:15 -04:00
Al
84cf21df88 [osm] Separating address formatter into its own module, adding some documentation of the various training sets with examples 2015-09-20 20:05:46 -04:00
Al
6731395ca0 [osm] Separating tagged from untagged output 2015-09-19 14:11:47 -04:00
Al
b85fe50fad [osm] Training data for toponyms only cares about valid languages for name field 2015-09-08 16:38:05 -07:00
Al
e566063343 [osm] Doing an all-to-nodes conversion and an additional filter on the borders data set 2015-09-08 09:18:08 -07:00
Al
8525529968 [osm] Not requiring qualified name tags to process OSM toponyms 2015-09-06 21:03:01 -07:00
Al
df20e2cbc0 [osm] Including toponyms in the training data for countries where the unqualified place names can be assumed to be examples of a given language 2015-09-04 14:13:33 -04:00
Al
17fcfa8b59 [fix] adding house to ignore keys rather than aliasing it 2015-09-04 12:40:08 -04:00
Al
d64a27bc57 [osm] Converting relations to nodes in borders training data 2015-09-04 12:32:25 -04:00
Al
168b7f59da [fix] default indices in strip_component 2015-09-04 12:29:47 -04:00
Al
64db63e3eb [osm] Removing house tag 2015-09-04 12:23:47 -04:00
Al
4ebdca0ea7 [fix] var 2015-09-03 21:01:20 -04:00
Al
8345afbcd0 [fix] exclude country toponyms where the default languages is well represented 2015-09-03 20:56:58 -04:00
Al
20bb191624 [fix] chaining 2015-09-03 20:52:00 -04:00
Al
e7cf5000fe [fix] Exclude polygons with > 1 regional language 2015-09-03 20:48:04 -04:00
Al
9a9530c1b9 [fix] unqualified names 2015-09-03 20:37:22 -04:00
Al
a5fdd911d8 [fix] only use name key for default names 2015-09-03 20:35:08 -04:00
Al
d8e1432533 [osm] Adding unqualified names in single-language countries 2015-09-03 20:31:49 -04:00
Al
b15d2d70aa [fix] top language 2015-09-03 20:09:46 -04:00
Al
44bf94a158 [osm] Better borders training data set (only need the metadata, not the polygons) 2015-09-03 20:09:03 -04:00
Al
55af9b0a0c [fix] OSM address tagged training data formatting 2015-09-03 18:35:19 -04:00
Al
c6bfc0e021 [osm] Postponing punctuation stripping until after address template rendering 2015-09-03 18:13:41 -04:00
Al
d54fb25e45 [osm] don't bother with the R-tree check if there are no name:* tags in border data set 2015-09-03 17:54:40 -04:00
Al
33af61095b [fix] var 2015-09-03 17:49:52 -04:00
Al
294101ad80 [osm] Treating components that are all punctuation as blank in address parsing (e.g. a single comma) 2015-09-03 17:46:57 -04:00
Al
e1e5c16637 [osm] Not adding unqualified name tags to toponym data set, throwing out a few cases of language ambiguity 2015-09-03 16:50:30 -04:00
Al
040a26a6f2 [fix] import 2015-09-03 13:54:23 -04:00
Al
7787427c58 [fix] typo 2015-09-03 13:53:18 -04:00
Al
23633e95dd [osm] Only adding country default language toponyms to training data 2015-09-03 13:44:41 -04:00
Al
11c01f64d2 [osm] OrderedDict of attrs in OSM training data 2015-09-03 11:11:18 -04:00
Al
27eb4e4aed [osm] Adding a toponym language training set using planet-borders.osm (all admin borders) 2015-09-03 10:19:11 -04:00
Al
db57855c95 [osm] Switching formatter repo to the OpenVenues fork, with fixes and several dozen new countries added 2015-09-03 10:06:54 -04:00
Al
a2ec8001b0 [osm] Removing postal code keys in formatted language training data 2015-08-24 14:08:36 -04:00
Al
e70c2453ee [fix] import 2015-08-22 15:04:30 -04:00
Al
3902715258 [osm] Some countries like Lebanon in OSM will list the same address under two languages (French/English), which creates an unreasonable task for a linear classifier, so running disambiguation in those cases 2015-08-22 14:11:49 -04:00
Al
4976be64e5 [fix] var name 2015-08-21 08:02:26 -04:00
Al
8e56568cab [fix] typo 2015-08-21 08:01:49 -04:00
Al
ca6d802a43 [languages] Moving language id methods into a separate package 2015-08-21 08:00:56 -04:00
Al
9d2f7e4bd1 [fix] var name 2015-08-18 16:20:12 -04:00
Al
0528d1b578 [osm] OSM untagged formatted addresses try to use language namespaced tags 2015-08-18 16:18:27 -04:00
Al
c09cb4dd82 [osm] OSM untagged formatted addresses now use the new language labeling scheme 2015-08-18 15:13:10 -04:00
Al
3daba2ddcd [fix] removing debug print 2015-08-18 13:22:48 -04:00
Al
ffe76f0403 [languages/osm] Checking for existence of separable prefix/suffix in the given dictionaries 2015-08-18 12:10:06 -04:00
Al
0e00625dbd [languages/osm] Adding a primitive phrase dictionary to the OSM training data construction script and a few heuristics to help disambiguate in the case of small local language groups that may not be specified with name:lang tags e.g. Occitan, Catalan, Basque, Galician, etc. Also throwing away ambiguous multilanguage names 2015-08-18 11:12:27 -04:00
Al
89071ea21a [osm] Omitting country in limited address data set (often abbreviated, doesn't convey language as well) 2015-08-15 03:25:45 -04:00
Al
c505260912 [fix] var name 2015-08-15 02:47:31 -04:00
Al
548ce79b99 [fix] street addresses by language 2015-08-15 02:44:04 -04:00
Al
74a751ce0a [osm] Adding a new OSM training data option for writing out full formatted addresses without place names 2015-08-15 02:39:49 -04:00
Al
0e92abd53e [osm] Adding building tag to venues training set construction 2015-08-14 21:07:07 -04:00