only consider partials in multi-words for initial count

This ensures that it is less likely that we exclude meaningful
words like 'hauptstrasse' just because they are frequent.

diff --git a/nominatim/tokenizer/legacy_icu_tokenizer.py b/nominatim/tokenizer/legacy_icu_tokenizer.py
index 5f83b73dc2c7c7fb6e7c444404549abf11c863cb..6bf409cca3ab3674b41605b06e8dfe49eda40e41 100644 (file)
--- a/nominatim/tokenizer/legacy_icu_tokenizer.py
+++ b/nominatim/tokenizer/legacy_icu_tokenizer.py
@@ -168,7 +168,8 @@ class LegacyICUTokenizer:
                 for name, cnt in cur:
                     terms = set()
                     for word in name_proc.get_variants_ascii(name_proc.get_normalized(name)):
-                        terms.update(word.split())
+                        if ' ' in word:
+                            terms.update(word.split())
                     for term in terms:
                         words[term] += cnt
 

diff --git a/test/python/test_tokenizer_legacy_icu.py b/test/python/test_tokenizer_legacy_icu.py
index 56c08e5aa11b5d623bb21daafea291e8e6f8f133..39fc9fb4c5a7f348c29ffe8c3b490caf458063f4 100644 (file)
--- a/test/python/test_tokenizer_legacy_icu.py
+++ b/test/python/test_tokenizer_legacy_icu.py
@@ -150,9 +150,8 @@ def test_init_word_table(tokenizer_factory, test_config, place_row, word_table):
     tok = tokenizer_factory()
     tok.init_new_db(test_config)
 
-    assert word_table.get_partial_words() == {('test', 1), ('52', 1),
+    assert word_table.get_partial_words() == {('test', 1),
                                               ('no', 1), ('area', 2),
-                                              ('holzstrasse', 1), ('holzstr', 1),
                                               ('holz', 1), ('strasse', 1),
                                               ('str', 1)}