update resnet example to contain #139

examples/ResNet/imagenet-resnet.py

@@ -86,25 +86,31 @@ def get_config(fake=False, data_format='NCHW'):
 
     if fake:
         logger.info("For benchmark, batch size is fixed to 64 per tower.")
-        dataset_train = dataset_val = FakeData(
+        dataset_train = FakeData(
             [[64, 224, 224, 3], [64]], 1000, random=False, dtype='uint8')
+        callbacks = []
     else:
         logger.info("Running on {} towers. Batch size per tower: {}".format(nr_tower, BATCH_SIZE))
         dataset_train = get_data('train')
         dataset_val = get_data('val')
-
-    return TrainConfig(
-        model=Model(data_format=data_format),
-        dataflow=dataset_train,
-        callbacks=[
+        callbacks = [
             ModelSaver(),
-            InferenceRunner(dataset_val, [
-                ClassificationError('wrong-top1', 'val-error-top1'),
-                ClassificationError('wrong-top5', 'val-error-top5')]),
             ScheduledHyperParamSetter('learning_rate',
                                       [(30, 1e-2), (60, 1e-3), (85, 1e-4), (95, 1e-5), (105, 1e-6)]),
             HumanHyperParamSetter('learning_rate'),
-        ],
+        ]
+        infs = [ClassificationError('wrong-top1', 'val-error-top1'),
+                ClassificationError('wrong-top5', 'val-error-top5')]
+        if nr_tower == 1:
+            callbacks.append(InferenceRunner(QueueInput(dataset_val), infs))
+        else:
+            callbacks.append(DataParallelInferenceRunner(
+                dataset_val, infs, list(range(nr_tower))))
+
+    return TrainConfig(
+        model=Model(data_format=data_format),
+        dataflow=dataset_train,
+        callbacks=callbacks,
         steps_per_epoch=5000,
         max_epoch=110,
         nr_tower=nr_tower

tensorpack/callbacks/inference_runner.py

@@ -78,6 +78,7 @@ class InferenceRunnerBase(Callback):
             self._size = input.size()
         except NotImplementedError:
             raise ValueError("Input used in InferenceRunner must have a size!")
+        logger.info("InferenceRunner will eval on an InputSource of size {}".format(self._size))
 
         if extra_hooks is None:
             extra_hooks = []

tensorpack/dataflow/common.py

@@ -137,8 +137,11 @@ class BatchData(ProxyDataFlow):
                     raise
                 except:
                     logger.exception("Cannot batch data. Perhaps they are of inconsistent shape?")
-                    import IPython as IP
-                    IP.embed(config=IP.terminal.ipapp.load_default_config())
+                    try:
+                        # open an ipython shell if possible
+                        import IPython as IP; IP.embed()    # noqa
+                    except:
+                        pass
         return result
 
 
@@ -679,14 +682,14 @@ class PrintData(ProxyDataFlow):
         """
         Dump gathered debugging information to stdout.
         """
-        msg = [""]
+        label = "" if self.name is None else " (" + self.label + ")"
+        logger.info(colored("DataFlow Info%s:" % label, 'cyan'))
         for i, dummy in enumerate(itertools.islice(self.ds.get_data(), self.num)):
             if isinstance(dummy, list):
-                msg.append("datapoint %i<%i with %i components consists of" % (i, self.num, len(dummy)))
+                msg = "datapoint %i<%i with %i components consists of\n" % (i, self.num, len(dummy))
                 for k, entry in enumerate(dummy):
-                    msg.append(self._analyze_input_data(entry, k))
-        label = "" if self.name is None else " (" + self.label + ")"
-        logger.info(colored("DataFlow Info%s:" % label, 'cyan') + '\n'.join(msg))
+                    msg += self._analyze_input_data(entry, k) + '\n'
+                print(msg)
 
         # reset again after print
         self.ds.reset_state()