Merge pull request #45 from beringresearch/multi-label-supervision

Multi label supervision
beringresearch · Sep 5, 2019 · 8a9886f · 8a9886f
2 parents 8aa67db + 4b475de
commit 8a9886f
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 5 deletions.
diff --git a/R-package/DESCRIPTION b/R-package/DESCRIPTION
@@ -1,6 +1,6 @@
 Package: ivis
 Title: Artificial neural network-driven visualization of high-dimensional data using triplets. 
-Version: 1.4.0
+Version: 1.4.1
 Authors@R: c(person("Benjamin", "Szubert", email = "[email protected]", role = c("aut", "cre")),
          person("Ignat", "Drozdov", email = "[email protected]", role = c("aut")),
          person("Kevin", "Rue-Albrecht", role = "ctb", email = "[email protected]", comment = c(ORCID = "0000-0003-3899-3872")))

diff --git a/ivis/ivis.py b/ivis/ivis.py
@@ -173,11 +173,19 @@ def _fit(self, X, Y=None, shuffle_mode=True):
                     if not is_multiclass(self.supervision_metric):
                         if not is_hinge(self.supervision_metric):
                             # Binary logistic classifier
-                            supervised_output = Dense(1, activation='sigmoid',
+                            if len(Y.shape) > 1:
+                                n_classes = Y.shape[-1]
+                            else:
+                                n_classes = 1
+                            supervised_output = Dense(n_classes, activation='sigmoid',
                                                       name='supervised')(anchor_embedding)
                         else:
                             # Binary Linear SVM output
-                            supervised_output = Dense(1, activation='linear',
+                            if len(Y.shape) > 1:
+                                n_classes = Y.shape[-1]
+                            else:
+                                n_classes = 1
+                            supervised_output = Dense(n_classes, activation='linear',
                                                       name='supervised',
                                                       kernel_regularizer=regularizers.l2())(anchor_embedding)
                     else:
@@ -193,7 +201,11 @@ def _fit(self, X, Y=None, shuffle_mode=True):
                                                       kernel_regularizer=regularizers.l2())(anchor_embedding)
                 else:
                     # Regression
-                    supervised_output = Dense(1, activation='linear',
+                    if len(Y.shape) > 1:
+                        n_classes = Y.shape[-1]
+                    else:
+                        n_classes = 1
+                    supervised_output = Dense(n_classes, activation='linear',
                                               name='supervised')(anchor_embedding)
 
                 final_network = Model(inputs=self.model_.inputs,

diff --git a/ivis/version.py b/ivis/version.py
@@ -1 +1 @@
-VERSION = '1.4.0'
+VERSION = '1.4.1'