!37815 modify the documents for op ScatterNd

Merge pull request !37815 from DeshiChen/code_docs_scatternd_master

docs/api/api_python/ops/mindspore.ops.func_scatter_nd.rst

@@ -5,13 +5,17 @@ mindspore.ops.scatter_nd
 
     根据指定的索引将更新值散布到新Tensor上。
 
-    使用给定的 `shape` 创建一个空Tensor，并将 `updates` 的值通过索引来设置空Tensor的值。空Tensor的秩为 :math:`P` ，而 `indices` 的秩为 :math:`Q` ， 且 :math:`Q \ge 2` 。
+    使用给定的 `shape` 创建一个空Tensor，并将 `updates` 的值通过索引（ `indices` ）来设置空Tensor的值。空Tensor的秩为 :math:`P` ， `indices` 的秩为 :math:`Q` 。
 
-    `indices` 的shape为 :math:`(i_0, i_1, ..., i_{Q-2}, N)` ， :math:`N \le P` 。
+    `shape` 为 :math:`(s_0, s_1, ..., s_{P-1})` ， 其中 :math:`P \ge 1` 。
 
-    `indices` 的最后一个维度（长度为 :math:`N` ）表示沿着空Tensor的 :math:`N` 个维度进行切片。
+    `indices` 的shape为 :math:`(i_0, i_1, ..., i_{Q-2}, N)` ， 其中 :math:`Q \ge 2` 且 :math:`N \le P` 。
 
-    `updates` 表示秩为 :math:`Q-1+P-N` 的Tensor，shape为 :math:`(i_0, i_1, ..., i_{Q-2}, shape_N, ..., shape_{P-1})` 。
+    `indices` 的最后一个维度（长度为 :math:`N` ）表示沿着空Tensor的第 :math:`N` 个维度进行切片。
+
+    `updates` 是一个秩为 :math:`Q-1+P-N` 的Tensor，shape为 :math:`(i_0, i_1, ..., i_{Q-2}, s_N, s_{N+1}, ..., s_{P-1})` 。
+
+    如果 `indices` 包含重复的下标，则 `updates` 的值会被累加到同一个位置上。
 
     在秩为3的第一个维度中插入两个新值矩阵的计算过程如下图所示：
 
@@ -19,9 +23,9 @@ mindspore.ops.scatter_nd
 
     **参数：**
 
-    - **indices** (Tensor) - 指定新Tensor中散布的索引，数据类型为int32或int64。索引的秩须至少为2，并且 :math:`indices\_shape[-1] \le len(shape)` 。
-    - **updates** (Tensor) - 指定更新Tensor，shape为 :math:`indices\_shape[:-1] + shape[indices\_shape[-1]:]` 。
-    - **shape** (tuple[int]) - 指定输出Tensor的shape，数据类型与索引相同。 `shape` 为 :math:`(x_1, x_2, ..., x_R)` 且 `shape` 的长度大于或等于2。换句话说， `shape` 至少是 :math:`(x_1, x_2)` ，且 `shape` 中的任何元素的值都必须大于等于1。也就是说， :math:`x_1 \ge 1`， :math:`x_2 \ge 1`。
+    - **indices** (Tensor) - 指定新Tensor中散布的索引，数据类型为int32或int64。索引的秩须至少为2，并且 `indices.shape[-1] <= len(shape)` 。
+    - **updates** (Tensor) - 指定更新Tensor，shape为 `indices.shape[:-1] + shape[indices.shape[-1]:]` 。
+    - **shape** (tuple[int]) - 指定输出Tensor的shape，数据类型与 `indices` 相同。 `shape` 不能为空，且 `shape` 中的任何元素的值都必须大于等于1。
 
     **返回：**
 

mindspore/python/mindspore/ops/function/array_func.py

@@ -1400,31 +1400,32 @@ def scatter_nd(indices, updates, shape):
     Scatters a tensor into a new tensor depending on the specified indices.
 
     Creates an empty tensor with the given `shape`, and set values by scattering the update tensor
-    depending on indices. The empty tensor has rank :math:`P` and `indices` has rank :math:`Q` where :math:`Q \ge 2`.
+    depending on indices. The empty tensor has rank :math:`P` and `indices` has rank :math:`Q`.
 
-    `indices` has shape :math:`(i_0, i_1, ..., i_{Q-2}, N)` where :math:`N \le P`.
+    The `shape` is :math:`(s_0, s_1, ..., s_{P-1})`, where :math:`P \ge 1`.
+
+    `indices` has shape :math:`(i_0, i_1, ..., i_{Q-2}, N)`, where :math:`Q \ge 2` and :math:`N \le P`.
 
     The last dimension of `indices` (with length :math:`N` ) indicates slices along the :math:`N` th dimension of the
     empty tensor.
 
-    `updates` is a tensor of rank :math:`Q-1+P-N`,
-    its shape is: :math:`(i_0, i_1, ..., i_{Q-2}, shape_N, ..., shape_{P-1})`.
+    `updates` is a tensor of rank :math:`Q-1+P-N`, and
+    its shape is :math:`(i_0, i_1, ..., i_{Q-2}, s_N, s_{N+1}, ..., s_{P-1})`.
 
-    The following figure shows the calculation process of inserting two slices in the first dimension of a rank-3
-    with two matrices of new values:
+    If `indices` contains duplicates, the duplicate `updates` are summed.
+
+    The following figure shows the calculation process of inserting two new value matrices into the first dimension
+    with rank-3:
 
     .. image:: ScatterNd.png
 
     Args:
-        indices (Tensor): The index of scattering in the new tensor with int32 or int64 data type.
-            The rank of indices must be at least 2 and :math:`indices\_shape[-1] \le len(shape)`.
-        updates (Tensor): The source Tensor to be scattered.
-            It has shape :math:`indices\_shape[:-1] + shape[indices\_shape[-1]:]`.
+        indices (Tensor): Define the index of scattering in the new tensor with int32 or int64 data type.
+            The rank of `indices` must be at least 2 and `indices.shape[-1] <= len(shape)`.
+        updates (Tensor): Define the source Tensor to be updated.
+            It has shape `indices.shape[:-1] + shape[indices.shape[-1]:]`.
         shape (tuple[int]): Define the shape of the output tensor, has the same data type as indices.
-            The shape of `shape` is :math:`(x_1, x_2, ..., x_R)`, and the length of 'shape' is greater than
-            or equal to 2. In other words, the shape of `shape` is at least :math:`(x_1, x_2)`.
-            And the value of any element in `shape` must be greater than or equal to 1.
-            In other words, :math:`x_1 \ge 1`, :math:`x_2 \ge 1`.
+            `shape` can not be empty, and the elements in `shape` must be greater than or equal to 1.
 
     Returns:
         Tensor, the new tensor, has the same type as `update` and the same shape as `shape`.