[Doc] Fix spelling, references and update info on building docs (#3682)

* Fix ref to message-passing guide

* Fix pygments and spacing

* Update build documentation steps in README.md

* Use links

* Adjust parameters in SAGEConv docstring in same order as init

* Fix spelling error

* Change doc link

docs/README.md

@@ -3,6 +3,7 @@ DGL document and tutorial folder
 
 Requirements
 ------------
+You need to build DGL locally first (as described [here](https://docs.dgl.ai/install/index.html#install-from-source)), and ensure the following python packages are installed:
 * sphinx
 * sphinx-gallery
 * sphinx_rtd_theme

docs/source/api/python/dgl.function.rst

@@ -6,7 +6,7 @@ dgl.function
 ==================================
 
 This subpackage hosts all the **built-in functions** provided by DGL. Built-in functions
-are DGL's recommended way to express different types of ref:`guide-message-passing` computation
+are DGL's recommended way to express different types of :ref:`guide-message-passing` computation
 (i.e., via :func:`~dgl.DGLGraph.update_all`) or computing edge-wise features from
 node-wise features (i.e., via :func:`~dgl.DGLGraph.apply_edges`). Built-in functions
 describe the node-wise and edge-wise computation in a symbolic way without any
@@ -55,7 +55,7 @@ following user-defined function:
    def udf_max(nodes):
       return {'h_max' : th.max(nodes.mailbox['m'], 1)[0]}
 
-All binary message function supports **broadcasting**, a mechansim for extending element-wise
+All binary message function supports **broadcasting**, a mechanism for extending element-wise
 operations to tensor inputs with different shapes. DGL generally follows the standard
 broadcasting semantic by `NumPy <https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html>`_
 and `PyTorch <https://pytorch.org/docs/stable/notes/broadcasting.html>`_. Below are some

docs/source/guide/message-efficient.rst

@@ -59,5 +59,5 @@ respectively:
 The above two implementations are mathematically equivalent. The latter
 one is more efficient because it does not need to save feat_src and
 feat_dst on edges, which is not memory-efficient. Plus, addition could
-be optimized with DGL’s built-in function ``u_add_v``, which further
+be optimized with DGL’s built-in function :func:`~dgl.function.u_add_v`, which further
 speeds up computation and saves memory footprint.

docs/source/guide/training-edge.rst

@@ -10,7 +10,7 @@ case, you would like to have an *edge classification/regression* model.
 
 Here we generate a random graph for edge prediction as a demonstration.
 
-.. code:: ipython3
+.. code:: python
 
     src = np.random.randint(0, 100, 500)
     dst = np.random.randint(0, 100, 500)
@@ -328,5 +328,3 @@ file <https://github.com/dmlc/dgl/tree/master/examples/pytorch/gcmc>`__
 is called ``GCMCLayer``. The edge type predictor module is called
 ``BiDecoder``. Both of them are more complicated than the setting
 described here.
-
-

docs/source/guide_cn/training-edge.rst

@@ -9,7 +9,7 @@
 
 以下代码生成了一个随机图用于演示边分类/回归。
 
-.. code:: ipython3
+.. code:: python
 
     src = np.random.randint(0, 100, 500)
     dst = np.random.randint(0, 100, 500)

docs/source/guide_ko/training-edge.rst

@@ -9,7 +9,7 @@
 
 우선, 예제로 사용할 에지 예측을 위한 임의의 그래프를 만든다.
 
-.. code:: ipython3
+.. code:: python
 
     src = np.random.randint(0, 100, 500)
     dst = np.random.randint(0, 100, 500)

python/dgl/function/message.py

@@ -229,7 +229,7 @@ def src_mul_edge(src, edge, out):
 
     Notes
     -----
-    This function is deprecated. Please use u_mul_e instead.
+    This function is deprecated. Please use :func:`~dgl.function.u_mul_e` instead.
 
     Parameters
     ----------
@@ -254,7 +254,7 @@ def copy_src(src, out):
 
     Notes
     -----
-    This function is deprecated. Please use copy_u instead.
+    This function is deprecated. Please use :func:`~dgl.function.copy_u` instead.
 
     Parameters
     ----------
@@ -281,7 +281,7 @@ def copy_edge(edge, out):
 
     Notes
     -----
-    This function is deprecated. Please use copy_e instead.
+    This function is deprecated. Please use :func:`~dgl.function.copy_e` instead.
 
     Parameters
     ----------

python/dgl/nn/mxnet/conv/sageconv.py

@@ -41,10 +41,10 @@ class SAGEConv(nn.Block):
         are required to be the same.
     out_feats : int
         Output feature size; i.e, the number of dimensions of :math:`h_i^{(l+1)}`.
-    feat_drop : float
-        Dropout rate on features, default: ``0``.
     aggregator_type : str
         Aggregator type to use (``mean``, ``gcn``, ``pool``, ``lstm``).
+    feat_drop : float
+        Dropout rate on features, default: ``0``.
     bias : bool
         If True, adds a learnable bias to the output. Default: ``True``.
     norm : callable activation function/layer or None, optional

python/dgl/nn/pytorch/conv/sageconv.py

@@ -50,10 +50,10 @@ class SAGEConv(nn.Module):
         are required to be the same.
     out_feats : int
         Output feature size; i.e, the number of dimensions of :math:`h_i^{(l+1)}`.
-    feat_drop : float
-        Dropout rate on features, default: ``0``.
     aggregator_type : str
         Aggregator type to use (``mean``, ``gcn``, ``pool``, ``lstm``).
+    feat_drop : float
+        Dropout rate on features, default: ``0``.
     bias : bool
         If True, adds a learnable bias to the output. Default: ``True``.
     norm : callable activation function/layer or None, optional
@@ -199,7 +199,7 @@ class SAGEConv(nn.Module):
         torch.Tensor
             The output feature of shape :math:`(N_{dst}, D_{out})`
             where :math:`N_{dst}` is the number of destination nodes in the input graph,
-            math:`D_{out}` is size of output feature.
+            :math:`D_{out}` is the size of the output feature.
         """
         self._compatibility_check()
         with graph.local_scope():

python/dgl/nn/tensorflow/conv/sageconv.py

@@ -40,10 +40,10 @@ class SAGEConv(layers.Layer):
         are required to be the same.
     out_feats : int
         Output feature size; i.e, the number of dimensions of :math:`h_i^{(l+1)}`.
-    feat_drop : float
-        Dropout rate on features, default: ``0``.
     aggregator_type : str
         Aggregator type to use (``mean``, ``gcn``, ``pool``, ``lstm``).
+    feat_drop : float
+        Dropout rate on features, default: ``0``.
     bias : bool
         If True, adds a learnable bias to the output. Default: ``True``.
     norm : callable activation function/layer or None, optional