address more pr comments

Signed-off-by: Matthew Owen <mowen@anyscale.com>

doc/source/data/loading-data.rst

@@ -983,5 +983,5 @@ utilize the cluster, ranging from ``1...override_num_blocks`` tasks. In other wo
 the higher the ``override_num_blocks``, the smaller the data blocks in the Dataset and
 hence more opportunities for parallel execution.
 
-For more information on how to tune the number of output blocks, see
-:ref:`Tuning output blocks for read <read_output_blocks>`.
+For more information on how to tune the number of output blocks and other suggestions
+for optimizing read performance, see `Optimizing reads <performance-tips.html#optimizing-reads>`__.

doc/source/data/transforming-data.rst

@@ -179,9 +179,7 @@ The user defined function you pass to :meth:`~ray.data.Dataset.map_batches` is m
 in multiple ways (see :ref:`Configuring batch format <configure_batch_format>`), the function should be of type
 ``Callable[DataBatch, DataBatch]``, where ``DataBatch = Union[pd.DataFrame, Dict[str, np.ndarray]]``. In
 other words, your function should take as input and output a batch of data which you can represent as a
-pandas DataFrame or a dictionary with string keys and NumPy ndarrays values. The function doesn't need
-to return a batch in the same format as its input, so you could input a pandas DataFrame and output a
-dictionary of NumPy ndarrays. For example, your function might look like:
+pandas DataFrame or a dictionary with string keys and NumPy ndarrays values. For example, your function might look like:
 
 .. testcode::