add local_step as callback property. remove some legacy code.

tensorpack/callbacks/base.py

@@ -17,6 +17,7 @@ class Callback(object):
     Attributes:
         epoch_num(int): the number of the current epoch.
         global_step(int): the number of global steps that have finished.
+        local_step(int): the local steps within the current epoch.
         trainer(Trainer): the trainer.
         graph(tf.Graph): the graph.
 
@@ -157,6 +158,10 @@ class Callback(object):
     def global_step(self):
         return self.trainer.global_step
 
+    @property
+    def local_step(self):
+        return self.trainer.local_step
+
     def __str__(self):
         return type(self).__name__
 

tensorpack/callbacks/steps.py

@@ -49,9 +49,8 @@ class StepTensorPrinter(Callback):
 
 class MaintainStepCounter(Callback):
     """
-    It maintains the global step in the graph and also creates the local step tensor.
-    This callback is always enabled by the trainer, and you wouldn't need to
-    use it.
+    It maintains the global step in the graph, making sure it's increased by one in every `run_step` call.
+    This callback is always enabled by the trainer, and you wouldn't need to use it.
     """
     def _setup_graph(self):
         # ensure it exists
@@ -69,12 +68,12 @@ class MaintainStepCounter(Callback):
         gs_val = get_global_step_value()
         if gs_val != 0:
             logger.info("Start training with global_step={}".format(gs_val))
-        self._last_updated = self.trainer.local_step
+        self._last_updated = self.local_step
 
     def _before_run(self, _):
         # increase global_step, when trainer.local_step changed
-        if self.trainer.local_step != self._last_updated:
-            self._last_updated = self.trainer.local_step
+        if self.local_step != self._last_updated:
+            self._last_updated = self.local_step
             return self._fetches
         else:
             return None
@@ -93,7 +92,7 @@ class ProgressBar(Callback):
         self._tags = [get_op_tensor_name(n)[0].split("/")[-1] for n in names]
 
     def _before_train(self):
-        self._last_updated = self.trainer.local_step
+        self._last_updated = self.local_step
 
         self._total = self.trainer.config.steps_per_epoch
         self._tqdm_args = get_tqdm_kwargs(leave=True)
@@ -104,10 +103,11 @@ class ProgressBar(Callback):
             self._tqdm_args['bar_format'] = self._tqdm_args['bar_format'] + "{postfix} "
 
     def _before_run(self, _):
-        if self.trainer.local_step != self._last_updated:
-            self._last_updated = self.trainer.local_step
+        # update progress bar when local step changed (one step is finished)
+        if self.local_step != self._last_updated:
+            self._last_updated = self.local_step
 
-            if self.trainer.local_step == 0:
+            if self.local_step == 0:
                 self._bar = tqdm.trange(self._total, **self._tqdm_args)
 
             return self._fetches
@@ -121,7 +121,7 @@ class ProgressBar(Callback):
 
     def _trigger_step(self):
         self._bar.update()
-        if self.trainer.local_step == self._total - 1:
+        if self.local_step == self._total - 1:
             self._bar.close()
 
     def _after_train(self):

tensorpack/callbacks/summary.py

@@ -60,7 +60,7 @@ class MergeAllSummaries(Callback):
     def _before_run(self, ctx):
         if self._run_alone:
             return None
-        if self.trainer.local_step == self._total - 1:
+        if self.local_step == self._total - 1:
             return self._fetches
         return None
 

tensorpack/callbacks/trigger.py

@@ -12,7 +12,7 @@ __all__ = ['PeriodicTrigger', 'PeriodicCallback']
 
 class PeriodicTrigger(ProxyCallback):
     """
-    Schedule to trigger a callback every k steps or every k epochs by its ``_trigger()`` method.
+    Schedule to trigger a callback every k steps or every k epochs by its ``trigger()`` method.
     """
     def __init__(self, triggerable, every_k_steps=None, every_k_epochs=None):
         """
@@ -37,7 +37,7 @@ class PeriodicTrigger(ProxyCallback):
             return
         # trigger_step is triggered after run_step, so
         # local_step + 1 is the number of step that have finished
-        if (self.trainer.local_step + 1) % self._step_k == 0:
+        if (self.local_step + 1) % self._step_k == 0:
             self.cb.trigger()
 
     def _trigger_epoch(self):

tensorpack/train/base.py

@@ -16,13 +16,13 @@ from .predict import PredictorFactory
 from .config import TrainConfig
 from .monitor import Monitors, TrainingMonitor
 from ..utils import logger
-from ..utils.develop import deprecated, log_deprecated
+from ..utils.develop import deprecated
 from ..callbacks import Callback, Callbacks, MaintainStepCounter
 from ..tfutils import get_global_step_value
 from ..tfutils.model_utils import describe_model
 from ..tfutils.sesscreate import ReuseSessionCreator
 
-__all__ = ['Trainer', 'StopTraining', 'MultiPredictorTowerTrainer']
+__all__ = ['Trainer', 'StopTraining']
 
 
 class StopTraining(BaseException):
@@ -211,13 +211,3 @@ class Trainer(object):
     @deprecated("Use get_predictors instead!", "2017-05-20")
     def get_predict_funcs(self, input_names, output_names, n):
         return self.get_predictors(input_names, output_names, n)
-
-    @deprecated("Don't need to call it any more!", "2017-03-20")
-    def _setup_predictor_factory(self):
-        pass
-
-
-# back-compat
-class MultiPredictorTowerTrainer(Trainer):
-    def __init__(self, *args, **kwargs):
-        log_deprecated("MultiPredictorTowerTrainer", "Just remove it instead.", "2017-03-21")

tensorpack/train/config.py

@@ -162,17 +162,6 @@ class TrainConfig(object):
 
         assert len(kwargs) == 0, 'Unknown arguments: {}'.format(str(kwargs.keys()))
 
-    def set_tower(self, nr_tower=None, tower=None):
-        log_deprecated("config.set_tower", "Set config.tower or config.nr_tower directly.", "2017-03-15")
-        assert nr_tower is None or tower is None, "Cannot set both nr_tower and tower!"
-        if nr_tower:
-            tower = list(range(nr_tower))
-        else:
-            if isinstance(tower, int):
-                tower = list(range(tower))
-        self.tower = tower
-        assert isinstance(self.tower, list)
-
     @property
     def nr_tower(self):
         return len(self.tower)