fix saver v1/v2 issues

tensorpack/callbacks/common.py

@@ -75,16 +75,6 @@ due to an alternative in a different tower".format(v.name, var_dict[name].name))
                 self.path,
                 global_step=get_global_step(),
                 write_meta_graph=False)
-
-            # create a symbolic link for the latest model
-            latest = self.saver.last_checkpoints[-1]
-            basename = os.path.basename(latest)
-            linkname = os.path.join(os.path.dirname(latest), 'latest')
-            try:
-                os.unlink(linkname)
-            except OSError:
-                pass
-            os.symlink(basename, linkname)
         except (OSError, IOError):   # disk error sometimes.. just ignore it
             logger.exception("Exception in ModelSaver.trigger_epoch!")
 

tensorpack/models/model_desc.py

@@ -84,7 +84,6 @@ Use _build_graph(self, input_vars) and get_current_tower_context().is_training i
                 CheckGradient()
                 ]
 
-
 class ModelFromMetaGraph(ModelDesc):
     """
     Load the whole exact TF graph from a saved meta_graph.

tensorpack/tfutils/sessinit.py

@@ -54,27 +54,29 @@ class SaverRestore(SessionInit):
     """
     def __init__(self, model_path, prefix=None):
         """
-        :param model_path: a model file or a ``checkpoint`` file.
+        :param model_path: a model name (model-xxxx) or a ``checkpoint`` file.
         :param prefix: add a `prefix/` for every variable in this checkpoint
         """
-        assert os.path.isfile(model_path)
         if os.path.basename(model_path) == model_path:
             model_path = os.path.join('.', model_path)  # avoid #4921
         if os.path.basename(model_path) == 'checkpoint':
-            model_path = tf.train.get_checkpoint_state(
-                os.path.dirname(model_path)).model_checkpoint_path
-            assert os.path.isfile(model_path)
+            model_path = tf.train.latest_checkpoint(os.path.dirname(model_path))
+            # to be consistent with either v1 or v2
+        assert os.path.isfile(model_path) or os.path.isfile(model_path + '.index')
         self.set_path(model_path)
         self.prefix = prefix
 
     def _init(self, sess):
         logger.info(
-            "Restoring checkpoint from {}.".format(self.path))
+            "Restoring checkpoint from {} ...".format(self.path))
         chkpt_vars = SaverRestore._read_checkpoint_vars(self.path)
         vars_map = self._get_vars_to_restore_multimap(chkpt_vars)
         for dic in SaverRestore._produce_restore_dict(vars_map):
             # multiple saver under same name scope would cause error:
             # training/saver.py: assert restore_op.name.endswith("restore_all"), restore_op.name
+            try:
+                saver = tf.train.Saver(var_list=dic, name=str(id(dic)), write_version=2)
+            except:
                 saver = tf.train.Saver(var_list=dic, name=str(id(dic)))
             saver.restore(sess, self.path)
 

tensorpack/tfutils/symbolic_functions.py

@@ -10,7 +10,7 @@ def prediction_incorrect(logits, label, topk=1, name='incorrect_vector'):
     """
     :param logits: NxC
     :param label: N
-    :returns: a float32 vector of length N with 0/1 values, 1 meaning incorrect prediction
+    :returns: a float32 vector of length N with 0/1 values. 1 means incorrect prediction
     """
     return tf.cast(tf.logical_not(tf.nn.in_top_k(logits, label, topk)),
             tf.float32, name=name)
@@ -95,9 +95,7 @@ def rms(x, name=None):
             return tf.sqrt(tf.reduce_mean(tf.square(x)), name=name)
     return tf.sqrt(tf.reduce_mean(tf.square(x)), name=name)
 
-def huber_loss(x, delta=1, name=None):
-    if name is None:
-        name = 'huber_loss'
+def huber_loss(x, delta=1, name='huber_loss'):
     sqrcost = tf.square(x)
     abscost = tf.abs(x)
     return tf.reduce_sum(

tensorpack/train/trainer.py

@@ -78,7 +78,7 @@ class SimpleTrainer(Trainer):
 
         self.train_op = tf.group(
             self.config.optimizer.apply_gradients(grads, get_global_step_var()),
-            summary_moving_average())
+            summary_moving_average(), name='train_op')
 
         # create an infinte data producer
         self.config.dataset.reset_state()