Early stopping now uses the overall loss function instead of the intercomponent overlap

source/model_utils.py

@@ -275,7 +275,7 @@ class Entropy_regulariser(Layer):
     Identity layer that adds cluster weight entropy to the loss function
     """
 
-    def __init__(self, weight=1., *args, **kwargs):
+    def __init__(self, weight=0., *args, **kwargs):
         self.weight = weight
         super(Entropy_regulariser, self).__init__(*args, **kwargs)
 
@@ -290,7 +290,7 @@ class Entropy_regulariser(Layer):
         )
 
         # Adds metric that monitors dead neurons in the latent space
-        self.add_metric(-entropy, aggregation="mean", name="weight_entropy")
+        self.add_metric(entropy, aggregation="mean", name="-weight_entropy")
 
         self.add_loss(self.weight * K.sum(entropy), inputs=[z])