Improve deep

Author: antoinedemathelin

adapt/feature_based/_adda.py

@@ -407,7 +407,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys, disc_src, disc_tgt,
+    def get_loss(self, inputs_ys, inputs_yt, disc_src, disc_tgt,
                   disc_tgt_nograd, task_tgt):
 
         loss_disc = (-K.log(disc_src + EPS)

adapt/feature_based/_cdan.py

@@ -259,7 +259,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys,
+    def get_loss(self, inputs_ys, inputs_yt,
                  task_src, task_tgt,
                  disc_src, disc_tgt,
                  task_src_nograd,

adapt/feature_based/_dann.py

@@ -225,7 +225,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys,
+    def get_loss(self, inputs_ys, inputs_yt,
                   task_src, task_tgt,
                   disc_src, disc_tgt):
 

adapt/feature_based/_deep.py

@@ -244,14 +244,17 @@ def __init__(self,
         self.random_state = random_state
 
 
+    def _initialize_networks(self, shape_Xt):
+        zeros_enc_ = self.encoder_.predict(np.zeros((1,) + shape_Xt));
+        self.task_.predict(zeros_enc_);
+        self.discriminator_.predict(zeros_enc_);
+    
+    
     def _build(self, shape_Xs, shape_ys,
                     shape_Xt, shape_yt):
-        
         # Call predict to avoid strange behaviour with
         # Sequential model whith unspecified input_shape
-        zeros_enc_ = self.encoder_.predict(np.zeros((1,) + shape_Xt));
-        self.task_.predict(zeros_enc_);
-        self.discriminator_.predict(zeros_enc_);
+        self._initialize_networks(shape_Xt)
 
         inputs_Xs = Input(shape_Xs)
         inputs_ys = Input(shape_ys)
@@ -271,6 +274,7 @@ def _build(self, shape_Xs, shape_ys,
         self.model_ = Model(inputs, outputs)
 
         loss = self.get_loss(inputs_ys=inputs_ys,
+                             inputs_yt=inputs_yt,
                               **outputs)
         metrics = self.get_metrics(inputs_ys=inputs_ys,
                                     inputs_yt=inputs_yt,
@@ -390,14 +394,17 @@ def create_model(self, inputs_Xs, inputs_Xt):
         pass
 
 
-    def get_loss(self, inputs_ys, **ouputs):
+    def get_loss(self, inputs_ys, inputs_yt, **ouputs):
         """
         Get loss.
         
         Parameters
         ----------
         inputs_ys : InputLayer
             Input layer for ys entries.
+            
+        inputs_yt : InputLayer
+            Input layer for yt entries.
         
         outputs : dict of tf Tensors
             Model outputs tensors.

adapt/feature_based/_deepcoral.py

@@ -209,7 +209,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys,
+    def get_loss(self, inputs_ys, inputs_yt,
                   task_src, task_tgt,
                   cov_src, cov_tgt):
 

adapt/feature_based/_mcd.py

@@ -238,7 +238,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys, task_src,
+    def get_loss(self, inputs_ys, inputs_yt, task_src,
                   task_tgt, task_sec_src, task_sec_tgt):
 
         loss_task = 0.5 * (self.loss_(inputs_ys, task_src) + self.loss_(inputs_ys, task_sec_src))

adapt/feature_based/_mdd.py

@@ -171,7 +171,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys, task_src,
+    def get_loss(self, inputs_ys, inputs_yt, task_src,
                  task_src_nograd, task_tgt_nograd,
                  task_tgt, disc_src, disc_tgt):
 

adapt/feature_based/_wdgrl.py

@@ -225,7 +225,7 @@ def create_model(self, inputs_Xs, inputs_Xt):
         return outputs
 
 
-    def get_loss(self, inputs_ys,
+    def get_loss(self, inputs_ys, inputs_yt,
                   task_src, task_tgt,
                   disc_src, disc_tgt,
                   disc_grad):

adapt/utils.py

@@ -306,12 +306,17 @@ def get_default_encoder():
     return model
 
 
-def get_default_task():
+def get_default_task(activation=None):
     """
     Return a tensorflow Model of two hidden layers
     with 10 neurons each and relu activations. The
     last layer is composed of one neuron with linear
     activation.
+    
+    Parameters
+    ----------
+    activation : str (default=None)
+        Final activation
 
     Returns
     -------
@@ -321,7 +326,7 @@ def get_default_task():
     model.add(Flatten())
     model.add(Dense(10, activation="relu"))
     model.add(Dense(10, activation="relu"))
-    model.add(Dense(1, activation=None))
+    model.add(Dense(1, activation=activation))
     return model
 
 

tests/test_utils.py

@@ -50,7 +50,8 @@ def is_equal_estimator(v1, v2):
     elif isinstance(v1, Tree):
         pass # TODO create a function to check if two tree are equal
     else:
-        assert v1 == v2
+        if not "input" in str(v1):
+            assert v1 == v2
     return True