From 5605ba3185c5ed54cf36c6df52ddf06e3a4c621b Mon Sep 17 00:00:00 2001 From: Al Date: Thu, 6 Apr 2017 11:43:54 -0400 Subject: [PATCH] [docs] adding note about the newly-trained language classifier trained with FTRL-Proximal (now 1/10th the size), which keeps its high accuracy while maintaining a sparse solution. This commit will trigger a build with the freshly uploaded model. --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index c453f313..bc70438b 100644 --- a/README.md +++ b/README.md @@ -433,8 +433,8 @@ tagged traning examples for every inhabited country in the world. Many types of are performed to make the training data resemble real messy geocoder input as closely as possible. - **Language classification**: multinomial logistic regression -trained on all of OpenStreetMap ways, addr:* tags, toponyms and formatted -addresses. Labels are derived using point-in-polygon tests in Quattroshapes +trained (using the [FTRL-Proximal](https://research.google.com/pubs/archive/41159.pdf) method to induce sparsity) on all of OpenStreetMap ways, addr:* tags, toponyms and formatted +addresses. Labels are derived using point-in-polygon tests for both OSM countries and official/regional languages for countries and admin 1 boundaries respectively. So, for example, Spanish is the default language in Spain but in different regions e.g. Catalunya, Galicia, the Basque region, the respective