Make TensorFlow follow some of its own deprecation warnings

tensorflow/python/framework/func_graph.py

@@ -370,7 +370,6 @@ def _capture_by_value(
       name=None,
       attrs=None,
       op_def=None,
-      compute_shapes=True,
       compute_device=True):
     # When capturing by value, do the read outside
     reverse_captures = dict((v, k) for k, v in self.captures.items())
@@ -384,7 +383,7 @@ def _capture_by_value(
       else:
         op = ops.get_default_graph().create_op(
             op_type, uncaptured_inputs, dtypes, input_types, name, attrs,
-            op_def, compute_shapes, compute_device)
+            op_def, compute_device=compute_device)
         value = op.outputs[0]
     captured_value = self.capture(value)
     return captured_value.op
@@ -432,11 +431,12 @@ def create_op(
     Returns:
       An `Operation` object.
     """
+    del compute_shapes
     if self.capture_by_value and op_type in ["ReadVariableOp",
                                              "ResourceGather"]:
       return self._capture_by_value(
           op_type, inputs, dtypes, input_types, name, attrs, op_def,
-          compute_shapes, compute_device)
+          compute_device)
 
     # This capturing logic interacts poorly with control flow contexts which
     # want to replace inputs of ops far too late in the process. This can lead

tensorflow/python/framework/function.py

@@ -1019,8 +1019,7 @@ def _call(sig, *inputs, **kwargs):
       output_types,
       name=name,
       attrs=attrs,
-      op_def=sig,
-      compute_shapes=False)
+      op_def=sig)
   if op.outputs:
     if len(op.outputs) == 1:
       ret = op.outputs[0]

tensorflow/python/ops/functional_ops.py

@@ -853,7 +853,6 @@ def partitioned_call(args, f, tout=None, executing_eagerly=None, config=None,
       op_name,
       args,
       tout,
-      compute_shapes=False,
       name="PartitionedFunctionCall",
       attrs={
           "Tin": tin_attr,

tensorflow/python/ops/variable_scope.py

@@ -893,7 +893,7 @@ def _get_single_variable(self,
       else:
         # Instantiate initializer if provided initializer is a type object.
         if tf_inspect.isclass(initializer):
-          initializer = initializer(dtype=dtype)
+          initializer = initializer()
         if shape is not None and shape.is_fully_defined():
           init_val = lambda: initializer(  # pylint: disable=g-long-lambda
               shape.as_list(),

tensorflow/python/training/moving_averages.py

@@ -208,9 +208,7 @@ def _zero_debias(unbiased_var, value, decay):
                                               decay]) as scope:
     with ops.colocate_with(unbiased_var):
       with ops.init_scope():
-        biased_initializer = init_ops.zeros_initializer(
-            dtype=unbiased_var.dtype)(
-                unbiased_var.get_shape())
+        biased_initializer = init_ops.zeros_initializer()
         local_step_initializer = init_ops.zeros_initializer()
 
       def _maybe_get_unique(name):
@@ -230,8 +228,8 @@ def _maybe_get_unique(name):
         return name + ("_%d" % idx)
 
       biased_var = variable_scope.get_variable(
-          _maybe_get_unique("biased"),
-          initializer=biased_initializer,
+          _maybe_get_unique("biased"), initializer=biased_initializer,
+          shape=unbiased_var.get_shape(), dtype=unbiased_var.dtype,
           trainable=False)
       local_step = variable_scope.get_variable(
           _maybe_get_unique("local_step"),

tensorflow/python/training/slot_creator.py

@@ -184,7 +184,7 @@ def create_zeros_slot(primary, name, dtype=None, colocate_with_primary=True):
     dtype = primary.dtype
   slot_shape = primary.get_shape()
   if slot_shape.is_fully_defined():
-    initializer = init_ops.zeros_initializer(dtype)
+    initializer = init_ops.zeros_initializer()
     return create_slot_with_initializer(
         primary, initializer, slot_shape, dtype, name,
         colocate_with_primary=colocate_with_primary)