Address comments in #1776

Author: edwardjkim

doc/v2.rst

@@ -263,6 +263,8 @@ The follow serializer/deserializer classes have been renamed and/or moved:
 | ``sagemaker.predictor._JsonDeserializer``              | ``sagemaker.deserializers.JSONDeserializer``          |
 +--------------------------------------------------------+-------------------------------------------------------+
 
+``sagemaker.serializers.LibSVMSerializer`` has been added in v2.0.
+
 ``distributions``
 ~~~~~~~~~~~~~~~~~
 
@@ -325,6 +327,11 @@ TensorFlow Serving Predictor
 ``sagemaker.tensorflow.serving.Predictor`` has been renamed to :class:`sagemaker.tensorflow.model.TensorFlowPredictor`.
 (For the previous implementation of that class, see `Deprecate Legacy TensorFlow <#deprecate-legacy-tensorflow>`_).
 
+XGBoost Predictor
+~~~~~~~~~~~~~~~~~
+
+The default seriazlier of ``sagemaker.xgboost.model.XGBoostPredictor`` has been changed from ``NumpySerializer`` to ``LibSVMSerializer``.
+
 
 Inputs
 ------

src/sagemaker/serializers.py

@@ -54,7 +54,7 @@ def CONTENT_TYPE(self):
 
 
 class CSVSerializer(BaseSerializer):
-    """Searilize data of various formats to a CSV-formatted string."""
+    """Serialize data of various formats to a CSV-formatted string."""
 
     CONTENT_TYPE = "text/csv"
 
@@ -102,7 +102,7 @@ def _serialize_row(self, data):
             csv_writer.writerow(data)
             return csv_buffer.getvalue().rstrip("\r\n")
 
-        raise ValueError("Unable to handle input format: ", type(data))
+        raise ValueError("Unable to handle input format: %s" % type(data))
 
     def _is_sequence_like(self, data):
         """Returns true if obj is iterable and subscriptable."""
@@ -273,7 +273,14 @@ def serialize(self, data):
 
 
 class LibSVMSerializer(BaseSerializer):
-    """Searilize data of various formats to a LibSVM-formatted string."""
+    """Serialize data of various formats to a LibSVM-formatted string.
+
+    The data must already be in LIBSVM file format:
+    <label> <index1>:<value1> <index2>:<value2> ...
+
+    It is suitable for sparse datasets since it does not store zero-valued
+    features.
+    """
 
     CONTENT_TYPE = "text/libsvm"
 
@@ -293,4 +300,4 @@ def serialize(self, data):
         if hasattr(data, "read"):
             return data.read()
 
-        raise ValueError("Unable to handle input format: ", type(data))
+        raise ValueError("Unable to handle input format: %s" % type(data))