Calculate recommended items without scanning the entire database by

using pre-rated items and comparing against similar items that have
already been stored

ratings/tests/tests.py

@@ -3,7 +3,7 @@
 
 from ratings.models import RatedItem
 from ratings.tests.models import Food, Beverage, BeverageRating
-from ratings.utils import sim_euclidean_distance, sim_pearson_correlation, top_matches, recommendations, calculate_similar_items
+from ratings.utils import sim_euclidean_distance, sim_pearson_correlation, top_matches, recommendations, calculate_similar_items, recommended_items
 
 
 class BaseRatingsTestCase(TestCase):
@@ -351,3 +351,8 @@ def test_similar_items(self):
 
         other_for_food_a = self.food_a.ratings.similar_items()[0]
         self.assertEqual(top_for_food_a, other_for_food_a)
+
+    def test_recommended_items(self):
+        calculate_similar_items(RatedItem.objects.all())
+        result = recommended_items(RatedItem.objects.all(), self.user_g)
+        self.assertEqual(str(result), '[(3.6100310668021822, <Food: food_a>), (3.5313950341859761, <Food: food_f>), (2.9609998607242685, <Food: food_c>)]')

ratings/utils.py

@@ -211,3 +211,31 @@ def _store_top_matches(ratings_queryset, rated_queryset, num, is_gfk):
             if created or si.score != score:
                 si.score = score
                 si.save()
+
+def recommended_items(ratings_queryset, user):
+    from ratings.models import SimilarItem
+    scores = {}
+    total_sim = {}
+
+    for item in ratings_queryset.filter(user=user):
+
+        for similar_item in SimilarItem.objects.get_for_item(item.content_object):
+
+            actual = similar_item.similar_object
+            lookup_kwargs = ratings_queryset.model.lookup_kwargs(actual)
+            lookup_kwargs['user'] = user
+
+            if ratings_queryset.filter(**lookup_kwargs):
+                continue
+
+            scores.setdefault(actual, 0)
+            scores[actual] += similar_item.score * item.score
+
+            total_sim.setdefault(actual, 0)
+            total_sim[actual] += similar_item.score
+
+    rankings = [(score/total_sim[item], item) for item, score in scores.iteritems()]
+
+    rankings.sort()
+    rankings.reverse()
+    return rankings