reduce duplicate model code, better name

- Multiple inheritance so we don't need DannRnnModel which only changed
  what it inherited from.
- Create BasicModel not FcnModelBase in methods.py

methods.py

@@ -114,7 +114,7 @@ def create_optimizers(self):
         self.checkpoint_variables["opt"] = self.opt
 
     def create_model(self):
-        self.model = models.FcnModelBase(self.num_classes, self.domain_outputs)
+        self.model = models.BasicModel(self.num_classes, self.domain_outputs)
 
     def create_losses(self):
         self.task_loss = make_loss()
@@ -1028,7 +1028,7 @@ def mle_for_p_d_given_y(self):
         self.p_d_given_y = p_d_given_y
 
     def create_model(self):
-        self.model = models.FcnModelBase(self.num_classes, self.domain_outputs)
+        self.model = models.BasicModel(self.num_classes, self.domain_outputs)
 
     def compute_losses(self, x, task_y_true, domain_y_true, task_y_pred,
             domain_y_pred, fe_output, training):

models.py

@@ -162,8 +162,18 @@ def __init__(self, num_classes, num_domains, **kwargs):
         ])
 
 
-class DannModel(FcnModelBase):
-    """ DANN adds a gradient reversal layer before the domain classifier """
+class BasicModel(FcnModelBase):
+    """ Model without adaptation (i.e. no DANN) """
+    pass
+
+
+class DannModelBase:
+    """ DANN adds a gradient reversal layer before the domain classifier
+
+    Note: we don't inherit from FcnModelBase or any other specific model because
+    we want to support either FcnModelBase, RnnModelBase, etc. with multiple
+    inheritance.
+    """
     def __init__(self, num_classes, num_domains, global_step,
             total_steps, **kwargs):
         super().__init__(num_classes, num_domains, **kwargs)
@@ -175,7 +185,12 @@ def call_domain_classifier(self, fe, task, **kwargs):
         return self.domain_classifier(grl_output, **kwargs)
 
 
-class HeterogeneousDannModel(DannModel):
+class DannModel(DannModelBase, FcnModelBase):
+    """ Model with adaptation (i.e. with DANN) """
+    pass
+
+
+class HeterogeneousDannModel(DannModelBase, FcnModelBase):
     """ Heterogeneous DANN model has multiple feature extractors,
     very similar to DannSmoothModel() code except this has multiple FE's
     not multiple DC's """
@@ -219,7 +234,7 @@ def call_domain_classifier(self, fe, task, which_fe=None, **kwargs):
         return self.domain_classifier(grl_output, **kwargs)
 
 
-class SleepModel(DannModel):
+class SleepModel(DannModelBase, FcnModelBase):
     """ Sleep model is DANN but concatenating task classifier output (with stop
     gradient) with feature extractor output when fed to the domain classifier """
     def __init__(self, *args, **kwargs):
@@ -234,7 +249,7 @@ def call_domain_classifier(self, fe, task, **kwargs):
         return self.domain_classifier(domain_input, **kwargs)
 
 
-class DannSmoothModel(DannModel):
+class DannSmoothModel(DannModelBase, FcnModelBase):
     """ DANN Smooth model hs multiple domain classifiers,
     very similar to HeterogeneousDannModel() code except this has multiple DC's
     not multiple FE's """
@@ -319,7 +334,7 @@ def call(self, inputs, **kwargs):
 
 class RnnModelBase(ModelBase):
     """ RNN-based model - for R-DANN and VRADA """
-    def __init__(self, vrada, num_classes, num_domains, **kwargs):
+    def __init__(self, num_classes, num_domains, vrada, **kwargs):
         super().__init__(**kwargs)
         self.num_classes = num_classes
         self.num_domains = num_domains
@@ -348,26 +363,11 @@ def call(self, inputs, training=None, **kwargs):
         return task, domain, fe
 
 
-class DannRnnModel(RnnModelBase):
-    """ DannModel but for RnnModelBase not FcnModelBase """
-    def __init__(self, vrada, num_classes, num_domains, global_step,
-            total_steps, **kwargs):
-        super().__init__(vrada, num_classes, num_domains, **kwargs)
-        grl_schedule = DannGrlSchedule(total_steps)
-        self.flip_gradient = FlipGradient(global_step, grl_schedule)
-
-    def call_domain_classifier(self, fe, task, **kwargs):
-        grl_output = self.flip_gradient(fe, **kwargs)
-        return self.domain_classifier(grl_output, **kwargs)
-
-
-class VradaModel(DannRnnModel):
+class VradaModel(DannModelBase, RnnModelBase):
     def __init__(self, *args, **kwargs):
-        vrada = True
-        super().__init__(vrada, *args, **kwargs)
+        super().__init__(*args, vrada=True, **kwargs)
 
 
-class RDannModel(DannRnnModel):
+class RDannModel(DannModelBase, RnnModelBase):
     def __init__(self, *args, **kwargs):
-        vrada = False
-        super().__init__(vrada, *args, **kwargs)
+        super().__init__(*args, vrada=False, **kwargs)