Removed deprecated functions and classes

csrank/fate_ranking.py

@@ -237,33 +237,6 @@ def _create_set_layers(self, **kwargs):
         else:
             self.set_layer = None
 
-    @deprecated
-    def _construct_scoring_model(self, n_objects_test, X, **kwargs):
-        """
-        Construct a scoring model for prediction on the given number of objects.
-
-        The existing layers of the networks are used during the construction
-        and the weights are shared.
-
-        Parameters
-        ----------
-        n_objects_test : int
-            Number of objects for which to compute scores
-        """
-        input_layer_scorer = Input(shape=(n_objects_test,
-                                          self.n_object_features),
-                                   name="input_node")
-        if self.n_hidden_set_layers >= 1:
-            set_repr = self.set_layer(input_layer_scorer)
-        else:
-            set_repr = None
-
-        scores = self.join_input_layers(input_layer_scorer, set_repr,
-                                        n_layers=self.n_hidden_set_layers,
-                                        n_objects=n_objects_test)
-        scoring_model = Model(inputs=input_layer_scorer, outputs=scores)
-        return scoring_model
-
     @staticmethod
     def _bucket_frequencies(X, min_bucket_size=32):
         """

csrank/objectranking/cmp_net.py

@@ -24,7 +24,7 @@
     N_UNITS_DEFAULT_RANGES
 from csrank.objectranking.object_ranker import ObjectRanker
 from csrank.tunable import Tunable
-from csrank.util import deprecated, tunable_parameters_ranges
+from csrank.util import tunable_parameters_ranges
 from ..dataset_reader.objectranking.util import generate_complete_pairwise_dataset
 
 __all__ = ['CmpNet']
@@ -158,34 +158,6 @@ def predict_pair(self, a, b, **kwargs):
     def evaluate(self, X1_test, X2_test, Y_test, **kwd):
         return self.model.evaluate([X1_test, X2_test], Y_test, **kwd)
 
-    @deprecated
-    def get_scores(self, X1, **kwd):
-        assert X1.shape[1] == self.n_features
-        objects = np.arange(X1.shape[0])
-        feature_ranks = np.concatenate((X1, np.array([objects]).T), axis=1)
-        pairs = np.array(list(combinations(feature_ranks, 2)))
-
-        predictions = self.predict_pair(pairs[:, 0, :-1], pairs[:, 1, :-1], **kwd)
-        score_total = {key: [] for key in objects}
-        for objs, scores in zip(pairs[:, :, -1], predictions):
-            score_total[int(objs[0])].append(scores[0])
-            score_total[int(objs[1])].append(scores[1])
-        scores_borda = dict()
-        for key, value in score_total.items():
-            scores_borda[key] = np.mean(np.array(value))
-        scores_borda = np.array(sorted(scores_borda.items(), key=operator.itemgetter(0)))
-        scores = scores_borda[:, 1]
-        return scores
-
-    @deprecated
-    def predict_ranks(self, X, **kwd):
-        Y = []
-        for X1 in X:
-            scores = self.get_scores(X1, **kwd)
-            r1 = len(scores) - rankdata(scores)
-            Y.append(r1)
-        return np.array(Y)
-
     @classmethod
     def set_tunable_parameter_ranges(cls, param_ranges_dict):
         logger = logging.getLogger("CmpNet")

csrank/objectranking/feta_ranker.py

@@ -20,14 +20,14 @@
     N_HIDDEN_LAYERS
 from csrank.objectranking.object_ranker import ObjectRanker
 from csrank.tunable import Tunable
-from csrank.util import deprecated, tunable_parameters_ranges, tensorify
+from csrank.util import tunable_parameters_ranges, tensorify
 
 __all__ = ['FETANetwork']
 
 
 class FETANetwork(ObjectRanker, Tunable):
-    """ Create a FETA-network architecture for object ranking.
-
+    """
+    Create a FETA-network architecture for object ranking.
     Training and prediction complexity is quadratic in the number of objects.
 
     Parameters
@@ -355,22 +355,4 @@ def tunable_parameters(cls):
                 (BATCH_SIZE, BATCH_SIZE_DEFAULT_RANGE),
             ])
             if cls._use_early_stopping:
-                cls._tunable[EARLY_STOPPING_PATIENCE] = EARLY_STOPPING_PATIENCE_DEFAULT_RANGE
-
-    @deprecated
-    def get_scores(self, X1):
-        assert X1.shape[1] == self.n_features
-        objects = np.arange(X1.shape[0])
-        feature_ranks = np.concatenate((X1, np.array([objects]).T), axis=1)
-        pairs = np.array(list(combinations(feature_ranks, 2)))
-        predictions = self._predict_pair(pairs[:, 0, :-1], pairs[:, 1, :-1])
-        score_total = {key: [] for key in objects}
-        for objs, scores in zip(pairs[:, :, -1], predictions):
-            score_total[int(objs[0])].append(scores[0])
-            score_total[int(objs[1])].append(scores[1])
-        scores_borda = dict()
-        for key, value in score_total.items():
-            scores_borda[key] = np.mean(np.array(value))
-        scores_borda = np.array(sorted(scores_borda.items(), key=operator.itemgetter(0)))
-        scores = scores_borda[:, 1]
-        return scores
+                cls._tunable[EARLY_STOPPING_PATIENCE] = EARLY_STOPPING_PATIENCE_DEFAULT_RANGE