Add a ModelDescBase without single-cost assumptions

tensorpack/graph_builder/model_desc.py

@@ -12,7 +12,7 @@ from ..utils.argtools import memoized
 from .input_source_base import InputSource
 from ..models.regularize import regularize_cost_from_collection
 
-__all__ = ['InputDesc', 'ModelDesc']
+__all__ = ['InputDesc', 'ModelDesc', 'ModelDescBase']
 
 
 class InputDesc(
@@ -87,12 +87,10 @@ class InputDesc(
 
 
 @six.add_metaclass(ABCMeta)
-class ModelDesc(object):
-    """ Base class for a model description.
-    """
+class ModelDescBase(object):
+    """ Base class for a model description.  """
 
-# inputs:
-# TODO remove this method?
+    # TODO remove this method? Now mainly used in predict/
     @memoized
     def get_reused_placehdrs(self):
         """
@@ -151,11 +149,18 @@ class ModelDesc(object):
     def _build_graph(self, inputs):
         pass
 
+
+class ModelDesc(ModelDescBase):
+    """
+    A ModelDesc with single cost and single optimizers.
+    """
+
     def get_cost(self):
         """
         Return the cost tensor in the graph.
         Used by some of the tensorpack :class:`Trainer` which assumes single-cost models.
-        You can ignore this method if you use your own trainer with more than one cost.
+        You can ignore this method (or just use :class:`ModelDescBase`)
+        if you use your own trainer with more than one cost.
 
         It calls :meth:`ModelDesc._get_cost()` which by default returns
         ``self.cost``. You can override :meth:`_get_cost()` if needed.
@@ -175,7 +180,7 @@ class ModelDesc(object):
         """
         Return the optimizer used in the task.
         Used by some of the tensorpack :class:`Trainer` which assume single optimizer.
-        You can (and should) ignore this method if you use a custom trainer with more than one optimizers.
+        You should use :class:`ModelDescBase` if you use a custom trainer with more than one optimizers.
 
         Users of :class:`ModelDesc` will need to implement `_get_optimizer()`,
         which will only be called once per each model.
@@ -187,9 +192,3 @@ class ModelDesc(object):
 
     def _get_optimizer(self):
         raise NotImplementedError()
-
-    def get_gradient_processor(self):
-        return self._get_gradient_processor()
-
-    def _get_gradient_processor(self):
-        return []