En docs for graph

Author: Seventeen17

docs/graph/data_loader_en.md

@@ -0,0 +1,131 @@
+# Data objects
+
+GraphLearn describes the result of traversal and sampling as a data object; GraphLearn traversal and sampling are Batch operations, in which the number of neighbors/negative neighbors of a Batch can be equal or unequal, so sampling is divided into aligned and non-aligned sampling.
+
+The result of vertex traversal and aligned vertex sampling is `Nodes`, and the result of non-aligned vertex sampling is `SparseNodes`. Correspondingly, edge traversal and aligned edge sampling results in `Edges`, and non-aligned edge sampling results in `SparseEdges`. <br />
+
+## Dense data objects
+### `Nodes`
+
+```python
+@property
+def ids(self):
+""" vertex id, numpy.ndarray(int64) """
+
+@property
+def shape(self):
+""" vertex id's shape, (batch_size) / (batch_size, neighbor_count) """
+
+@property
+def int_attrs(self):
+""" attributes of type int, numpy.ndarray(int64), shape as [ids.shape, number of attributes of type int] """
+
+@property
+def float_attrs(self):
+""" properties of type float, numpy.ndarray(float32), shape is [ids.shape, number of properties of type float] """
+
+@property
+def string_attrs(self):
+""" attributes of type string, numpy.ndarray(string), shape is [ids.shape, number of attributes of type string] """
+
+@property
+def weights(self):
+""" weights, numpy.ndarray(float32), shape is ids.shape """
+
+@property
+def labels(self):
+""" labels, numpy.ndarray(int32), shape is ids.shape """ @property
+def ids(self):
+""" vertex id, numpy.ndarray(int64) """
+
+@property
+def shape(self):
+""" vertex id's shape, (batch_size) / (batch_size, neighbor_count) """
+
+@property
+def int_attrs(self):
+""" attributes of type int, numpy.ndarray(int64), shape as [ids.shape, number of attributes of type int] """
+
+@property
+def float_attrs(self):
+""" properties of type float, numpy.ndarray(float32), shape is [ids.shape, number of properties of type float] """
+
+@property
+def string_attrs(self):
+""" attributes of type string, numpy.ndarray(string), shape is [ids.shape, number of attributes of type string] """
+
+@property
+def weights(self):
+""" weights, numpy.ndarray(float32), shape is ids.shape """
+
+@property
+def labels(self):
+""" labels, numpy.ndarray(int32), shape as ids.shape """
+```
+
+### `Edges`
+The difference between the `Edges` interface and `Nodes` is that the `ids` interface has been removed and the following four interfaces have been added for accessing source and destination vertices.
+
+```python
+@property
+def src_nodes(self):
+""" source vertex Nodes object """
+
+@property
+def dst_nodes(self):
+""" destination vertex Nodes object """
+
+@property
+def src_ids(self):
+""" source vertex id, numpy.ndarray(int64) """
+
+@property
+def dst_ids(self):
+""" destination vertex id, numpy.ndarray(int64) """
+```
+
+Regarding the shape of `ids`, in vertex and edge traversal operations, the shape is one-dimensional and the size is the specified batch size. In sampling operations, the shape is two-dimensional and the size is [the one-dimensional expansion size of the input data, the current number of samples].
+
+## Sparse data object
+
+### `SparseNodes`
+`SparseNodes` is used to express the sparse neighbor vertices of a vertex, with the following additional interface relative to Nodes.
+
+```python
+@property
+def offsets(self):
+""" one-dimensional shape-shifting array: the number of neighbors per vertex """
+
+@property
+def dense_shape(self):
+""" tuples with 2 elements: the shape of the corresponding Dense Nodes """
+
+@property
+def indices(self):
+""" 2-dimensional array representing the position of each neighbor """
+
+def __next__(self):
+""" the traversal interface, traversing the vertices of each vertex's neighbors """
+  return Nodes
+```
+
+### `SparseEdges`
+``SparseEdges`` is used to express the sparse neighboring edges of a vertex, with the following additional interface relative to Edges.
+
+```python
+@property
+def offsets(self):
+""" one-dimensional shape-shifting array: the number of neighbors per vertex """
+
+@property
+def dense_shape(self):
+""" tuples with 2 elements: the shape of the corresponding Dense Edges """
+
+@property
+def indices(self):
+""" 2-dimensional array representing the position of each neighbor """
+
+def __next__(self):
+""" the traversal interface, traversing the edges of each vertex's neighbors """
+  return Edges
+```

docs/graph/data_object_en.md

@@ -133,37 +133,36 @@ item	item         ->         item	item
 2	3                           3	1
 3	3                           2	3
 1	4                           3	2
-                              3	3
-                              1	4
-                              4	1
+                                3	3
+                                1	4
+                                4	1
 ```
 
 - src_type和dst_type不一致，如边类型为("user", "item", "u2i")，当u2i为无向边时，在加载时实际上除了加载原始的u2i边之外，额外加载了一份i2u的反向边。<br />
 
 ```
 原始u2i数据                 加载到图中的u2i数据    +     加载到图中的i2u数据
-user	item            ->      user	item                 item	user
+user	item            ->     user	item                 item	user
 1	2                           1	2                        2	1
 2	1                           2	1                        1	2
-1	3                           1	3												 3	1
-2	3                           2	3												 3	2
-3	3                           3	3												 3	3
-1	4                           1	4												 4	1
+1	3                           1	3						 3	1
+2	3                           2	3						 3	2
+3	3                           3	3						 3	3
+1	4                           1	4						 4	1
 ```
 
 <br />在遍历时，同构的无向边和异构的无向边使用详见GSL文档。
 <br />在采样时，需要根据自己指定的meta-path选择合理的边方向，合理使用outV(从src到dst)和inV(从dst到src)。outV和inV接口详见GSL文档。
 
 <a name="OVdVh"></a>
 ### partition
-分布式场景下，即存在多个GraphLeanrn Server时，构图时会自动进行图的partition, 将图分布式存储。默认partiton是按照src_id % server数进行节点和边的分配。
+分布式场景下，即存在多个GraphLearn Server时，构图时会自动进行图的partition, 将图分布式存储。默认partition是按照src_id % server数进行节点和边的分配。
 
-<a name="HNiIP"></a>
 ## 初始化
 
 顶点与边添加完成后，需要调用初始化接口，完成从原始数据到内存索引的构建。初始化过程决定了图数据被Serving的情况，单机的还是分布式的。若为分布式的，还要区分Server Mode和Client-Server Mode。初始化完成后，便可对Graph对象进行操作了。<br />
 
-<a name="nGHkF"></a>
+
 
 ### 单机
 单机模式比较简单，表示该Graph对象Hold全部图数据。<br />