Refine docs of reader in Transformer by following comments

fluid/neural_machine_translation/transformer/reader.py

@@ -64,8 +64,7 @@ class Pool(object):
 class DataReader(object):
     """
     The data reader loads all data from files and produces batches of data
-    in the way corresponding to settings. See the doc of __init__ function
-    for more setting details.
+    in the way corresponding to settings.
 
     An example of returning a generator producing data batches whose data
     is shuffled in each pass and sorted in each pool:
@@ -86,29 +85,6 @@ class DataReader(object):
         unk_mark='<unk>',
         clip_last_batch=False).batch_generator
     ```
-    """
-
-    def __init__(self,
-                 src_vocab_fpath,
-                 trg_vocab_fpath,
-                 fpattern,
-                 batch_size,
-                 pool_size,
-                 sort_type=SortType.NONE,
-                 clip_last_batch=True,
-                 tar_fname=None,
-                 min_length=0,
-                 max_length=100,
-                 shuffle=True,
-                 shuffle_batch=False,
-                 use_token_batch=False,
-                 delimiter="\t",
-                 start_mark="<s>",
-                 end_mark="<e>",
-                 unk_mark="<unk>",
-                 seed=0):
-        """
-        Load all data from files and set the settings to make mini-batches.
 
     :param src_vocab_fpath: The path of vocabulary file of source language.
     :type src_vocab_fpath: basestring
@@ -154,6 +130,26 @@ class DataReader(object):
     :param seed: The seed for random.
     :type seed: int
     """
+
+    def __init__(self,
+                 src_vocab_fpath,
+                 trg_vocab_fpath,
+                 fpattern,
+                 batch_size,
+                 pool_size,
+                 sort_type=SortType.NONE,
+                 clip_last_batch=True,
+                 tar_fname=None,
+                 min_length=0,
+                 max_length=100,
+                 shuffle=True,
+                 shuffle_batch=False,
+                 use_token_batch=False,
+                 delimiter="\t",
+                 start_mark="<s>",
+                 end_mark="<e>",
+                 unk_mark="<unk>",
+                 seed=0):
         self._src_vocab = self.load_dict(src_vocab_fpath)
         self._only_src = True
         if trg_vocab_fpath is not None: