use 'dataset' as key

example_cifar10.py

@@ -113,7 +113,7 @@ def get_config():
     tf.scalar_summary('learning_rate', lr)
 
     return dict(
-        dataset_train=dataset_train,
+        dataset=dataset_train,
         optimizer=tf.train.AdamOptimizer(lr),
         callback=Callbacks([
             SummaryWriter(),

example_mnist.py

@@ -126,7 +126,7 @@ def get_config():
     tf.scalar_summary('learning_rate', lr)
 
     return dict(
-        dataset_train=dataset_train,
+        dataset=dataset_train,
         optimizer=tf.train.AdamOptimizer(lr),
         callback=Callbacks([
             SummaryWriter(),

tensorpack/infer.py

@@ -18,8 +18,8 @@ def start_infer(config):
     Args:
         config: a tensorpack config dictionary
     """
-    dataset_test = config['dataset_test']
-    assert isinstance(dataset_test, DataFlow), dataset_test.__class__
+    dataset = config['dataset']
+    assert isinstance(dataset, DataFlow), dataset.__class__
 
     # a tf.ConfigProto instance
     sess_config = config.get('session_config', None)
@@ -53,7 +53,7 @@ def start_infer(config):
 
     with sess.as_default():
         with timed_operation('running one batch'):
-            for dp in dataset_test.get_data():
+            for dp in dataset.get_data():
                 feed = dict(zip(input_vars, dp))
                 fetches = [cost_var] + output_vars
                 results = sess.run(fetches, feed_dict=feed)

tensorpack/train.py

@@ -39,8 +39,8 @@ def start_train(config):
     Args:
         config: a tensorpack config dictionary
     """
-    dataset_train = config['dataset_train']
-    assert isinstance(dataset_train, DataFlow), dataset_train.__class__
+    dataset = config['dataset']
+    assert isinstance(dataset, DataFlow), dataset.__class__
 
     # a tf.train.Optimizer instance
     optimizer = config['optimizer']
@@ -90,7 +90,7 @@ def start_train(config):
     # start training:
     coord = tf.train.Coordinator()
     # a thread that keeps filling the queue
-    input_th = EnqueueThread(sess, coord, enqueue_op, dataset_train)
+    input_th = EnqueueThread(sess, coord, enqueue_op, dataset)
     model_th = tf.train.start_queue_runners(
         sess=sess, coord=coord, daemon=True, start=True)
     input_th.start()