misc cleanups

docs/casestudies/index.rst

-Casestudies
---------------------
-
-.. toctree::
-  :maxdepth: 1
-
-
-  colorize

docs/conf.py

@@ -359,7 +359,7 @@ def process_signature(app, what, name, obj, options, signature,
         signature = re.sub('tensorflow', 'tf', signature)
 
         # add scope name to layer signatures:
-        if hasattr(obj, 'use_scope') and hasattr(obj, 'symbolic_function'):
+        if hasattr(obj, 'use_scope'):
             if obj.use_scope:
                 signature = signature[0] + 'variable_scope_name, ' + signature[1:]
             elif obj.use_scope is None:

examples/DynamicFilterNetwork/steering-filter.py

@@ -88,7 +88,7 @@ class OnlineTensorboardExport(Callback):
         canvas = cv2.resize(canvas[..., None] * 255, (0, 0), fx=10, fy=10)
 
         self.trainer.monitors.put_image('filter_export', canvas)
-        # # you might also want to write these images to disk (as in the casestudy from the docs)
+        # # you might also want to write these images to disk
         # cv2.imwrite("export/out%04i.jpg" % self.cc, canvas)
         # self.cc += 1
 

tensorpack/models/registry.py

@@ -103,8 +103,8 @@ def layer_register(
                     if use_scope is False:
                         logger.warn(
                             "Please call layer {} without the first scope name argument, "
-                            "or register the layer with use_scope=None to allow "
-                            "two calling methods.".format(func.__name__))
+                            "or register the layer with use_scope=None to allow calling it "
+                            "with scope names.".format(func.__name__))
                     name, inputs = args[0], args[1]
                     args = args[1:]  # actual positional args used to call func
                 else:
@@ -147,8 +147,8 @@ def layer_register(
                 outputs = func(*args, **actual_args)
             return outputs
 
-        wrapped_func.symbolic_function = func   # attribute to access the underlying function object
         wrapped_func.use_scope = use_scope
+        wrapped_func.__argscope_enabled__ = True
         _register(func.__name__, wrapped_func)
         return wrapped_func
 

tensorpack/tfutils/__init__.py

@@ -42,8 +42,8 @@ so that these names will be invisible under `tensorpack.` namespace.
 
 To use these utilities, users are expected to import them explicitly, e.g.:
 
-import tensorpack.tfutils.symbolic_functions as symbf
+import tensorpack.tfutils.sessinit as sessinit
 """
 __all__.extend(['sessinit', 'summary', 'optimizer',
-                'sesscreate', 'gradproc', 'varreplace', 'symbolic_functions',
-                'distributed', 'tower'])
+                'sesscreate', 'gradproc', 'varreplace',
+                'tower'])

tensorpack/tfutils/argscope.py

@@ -40,14 +40,8 @@ def argscope(layers, **kwargs):
     if not isinstance(layers, list):
         layers = [layers]
 
-    # def _check_args_exist(l):
-    #     args = inspect.getargspec(l).args
-    #     for k, v in six.iteritems(kwargs):
-    #         assert k in args, "No argument {} in {}".format(k, l.__name__)
-
     for l in layers:
-        assert hasattr(l, 'symbolic_function'), "{} is not a registered layer".format(l.__name__)
-        # _check_args_exist(l.symbolic_function)
+        assert hasattr(l, '__argscope_enabled__'), "Argscope not supported for {}".format(l)
 
     # need to deepcopy so that changes to new_scope does not affect outer scope
     new_scope = copy.deepcopy(get_arg_scope())
@@ -119,8 +113,7 @@ def enable_argscope_for_function(func, log_shape=True):
                              out_tensor_descr.shape.as_list()))
 
         return out_tensor
-    # argscope requires this property
-    wrapped_func.symbolic_function = None
+    wrapped_func.__argscope_enabled__ = True
     return wrapped_func
 
 

tensorpack/tfutils/model_utils.py

@@ -6,12 +6,12 @@ from ..compat import tfv1 as tf
 from tabulate import tabulate
 from termcolor import colored
 
+from .common import get_op_tensor_name
 from ..utils import logger
 
 __all__ = []
 
 
-# TODO should also describe model_variables
 def describe_trainable_vars():
     """
     Print a description of the current model parameters.
@@ -39,7 +39,7 @@ def describe_trainable_vars():
 
         total += ele
         total_bytes += ele * v.dtype.size
-        data.append([v.name, shape, ele, v.device, v.dtype.base_dtype.name])
+        data.append([get_op_tensor_name(v.name)[0], shape, ele, v.device, v.dtype.base_dtype.name])
     headers = ['name', 'shape', '#elements', 'device', 'dtype']
 
     dtypes = list(set([x[4] for x in data]))