fixed transform_mean() when dealing with only one VG;

modified default settings (new default: aggregation: mean, all_prob:True)

src/learning/transformers.py

@@ -490,6 +490,8 @@ class DocEmbedderList:
 
     def transform_mean(self, lX, tfidf):
         if len(self.embedders) == 1:
+            if self.embedders[0].requires_tfidf:
+                lX = tfidf
             return self.embedders[0].transform(lX)
 
         langs = sorted(lX.keys())

src/util/parser_options.py

@@ -29,7 +29,7 @@ parser.add_option("--l2", dest="l2", action='store_true',
 parser.add_option("--allprob", dest="allprob", action='store_true',
                   help="All views are generated as posterior probabilities. This affects the supervised and pretrained"
                        "embeddings, for which a calibrated classifier is generated, which generates the posteriors",
-                  default=False)
+                  default=True)
 
 parser.add_option("--feat-weight", dest="feat_weight",
                   help="Term weighting function to weight the averaged embeddings", type=str,  default='tfidf')
@@ -58,7 +58,7 @@ parser.add_option("-z", "--zscore", dest="zscore", action='store_true',
 
 parser.add_option("-a", "--agg", dest="agg", action='store_true',
                   help="Set aggregation function of the common Z-space to average (Default: concatenation)",
-                  default=False)
+                  default=True)
 
 # ------------------------------------------------------------------------------------