Merge pull request #12 from takuti/fix-10

Disallow to build fature-based recommender from empty vector

flurs/data/entity.py

@@ -3,7 +3,7 @@
 
 class Base(object):
 
-    def __init__(self, index, feature=np.array([0.])):
+    def __init__(self, index, feature=np.array([])):
         self.index = index
         self.feature = feature
 
@@ -36,7 +36,7 @@ def index_one_hot(self, dim):
 class User(Base):
 
     def __str__(self):
-        if len(self.feature) == 1 and self.feature[0] == 0.:
+        if len(self.feature) == 0:
             return 'User(index={})'.format(self.index)
         else:
             return 'User(index={}, feature={})'.format(self.index, self.feature)
@@ -45,15 +45,15 @@ def __str__(self):
 class Item(Base):
 
     def __str__(self):
-        if len(self.feature) == 1 and self.feature[0] == 0.:
+        if len(self.feature) == 0:
             return 'Item(index={})'.format(self.index)
         else:
             return 'Item(index={}, feature={})'.format(self.index, self.feature)
 
 
 class Event(object):
 
-    def __init__(self, user, item, value=1., context=np.array([0.])):
+    def __init__(self, user, item, value=1., context=np.array([])):
         self.user = user
         self.item = item
         self.value = value
@@ -73,10 +73,12 @@ def encode(self, n_user=None, n_item=None,
                               feature=feature, vertical=False)
         x = np.concatenate((x, iv))
 
+        assert len(x) > 0, 'feature vector has zero dimension'
+
         return x if not vertical else np.array([x]).T
 
     def __str__(self):
-        if len(self.context) == 1 and self.context[0] == 0.:
+        if len(self.context) == 0:
             return 'Event(user={}, item={}, value={})'.format(self.user, self.item, self.value)
         else:
             return 'Event(user={}, item={}, value={}, context={})'.format(self.user, self.item, self.value, self.context)

flurs/recommender/factorization_machine.py

@@ -20,6 +20,16 @@ class FMRecommender(FactorizationMachine, FeatureRecommenderMixin):
     """
 
     def initialize(self, static=False, use_index=False):
+        """Initialize a recommender.
+
+        Parameters
+        ----------
+        static : bool, default=False
+            Disable incremental update if True.
+
+        use_index : bool, default=False
+            Incorporate onehot-encoded user/item index into a feature vector.
+        """
         super(FMRecommender, self).initialize()
         self.static = static
         self.use_index = use_index
@@ -111,7 +121,7 @@ def update(self, e, batch_train=False):
 
         self.update_model(x, e.value)
 
-    def score(self, user, candidates, context):
+    def score(self, user, candidates, context=np.array([])):
         # i_mat is (n_item_context, n_item) for all possible items
         # extract only target items
         i_mat = self.i_mat[:, candidates]

flurs/recommender/online_sketch.py

@@ -40,7 +40,7 @@ def update(self, e, batch_train=False):
         y = e.encode(index=False, feature=True, context=True)
         self.update_model(y)
 
-    def score(self, user, candidates, context):
+    def score(self, user, candidates, context=np.array([])):
         # i_mat is (n_item_context, n_item) for all possible items
         # extract only target items
         i_mat = self.i_mat[:, candidates]

flurs/recommender/tests/test_fm.py

@@ -23,13 +23,19 @@ def test_register_item(self):
     def test_update(self):
         self.recommender.register(User(0))
         self.recommender.register(Item(0))
-        self.recommender.update(Event(User(0), Item(0), 1))
+        self.recommender.update(
+            Event(User(0), Item(0), 1, context=np.array([1, 2, 3]))
+        )
         self.assertEqual(self.recommender.n_user, 1)
         self.assertEqual(self.recommender.n_item, 1)
 
     def test_score(self):
         self.recommender.register(User(0))
         self.recommender.register(Item(0))
-        self.recommender.update(Event(User(0), Item(0), 1))
-        score = self.recommender.score(User(0), np.array([0]), np.array([0]))
+        self.recommender.update(
+            Event(User(0), Item(0), 1, context=np.array([1, 2, 3]))
+        )
+        score = self.recommender.score(
+            User(0), candidates=np.array([0]), context=np.array([1, 2, 3])
+        )
         self.assertTrue(score >= 0.)

flurs/recommender/tests/test_sketch.py

@@ -22,13 +22,19 @@ def test_register_item(self):
     def test_update(self):
         self.recommender.register(User(0))
         self.recommender.register(Item(0))
-        self.recommender.update(Event(User(0), Item(0), 1))
+        self.recommender.update(
+            Event(User(0), Item(0), 1, context=np.array([1, 2, 3]))
+        )
         self.assertEqual(self.recommender.n_user, 1)
         self.assertEqual(self.recommender.n_item, 1)
 
     def test_score(self):
         self.recommender.register(User(0))
         self.recommender.register(Item(0))
-        self.recommender.update(Event(User(0), Item(0), 1))
-        score = self.recommender.score(User(0), np.array([0]), np.array([0]))
+        self.recommender.update(
+            Event(User(0), Item(0), 1, context=np.array([1, 2, 3]))
+        )
+        score = self.recommender.score(
+            User(0), candidates=np.array([0]), context=np.array([1, 2, 3])
+        )
         self.assertTrue(score >= 0. and score <= 1.0)