mindspore_gl.Graph

class mindspore_gl.Graph[source]

Graph class.

This is the class which should be annotated in construct function for GNNCell class.

adj_to_dense()[source]

Get the dense adjacent matrix of the graph.

Note

You must set vertex attr first due to the current limits of our system.

Returns: Tensor, a tensor with shape \((N, N)\), \(N\) is the number of nodes of the graph.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
...
>>> class TestAdjToDense(GNNCell):
...     def construct(self, g: Graph):
...         return g.adj_to_dense()
...
>>> ret = TestAdjToDense()(*graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[0, 1, 0, 0, 0, 0, 0, 0, 0],
     [0, 0, 0, 0, 0, 0, 0, 0, 0],
     [1, 1, 0, 0, 0, 0, 0, 0, 0],
     [0, 0, 0, 0, 0, 1, 0, 0, 0],
     [0, 0, 0, 1, 0, 0, 0, 0, 0],
     [0, 0, 0, 0, 1, 0, 1, 0, 0],
     [0, 0, 0, 0, 1, 0, 0, 0, 0],
     [0, 0, 0, 0, 0, 0, 0, 0, 0],
     [0, 0, 0, 0, 0, 0, 0, 0, 3]]

avg(neigh_feat)[source]

Aggregating node features from their neighbour and generates a node-level representation by aggregate function ‘avg’.

Parameters: neigh_feat (List[SrcVertex feature or Edge feature]) – a list of SrcVertex or Edge attributes represents the neighbour nodes or edges feature, with shape \((N, F)\), \(N\) is the number of SrcVertex or Edge, \(F\) is the feature dimension of the SrcVertex or Edge attribute.
Returns: Tensor, a tensor with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.
Raises: TypeError – If neigh_feat is not a list of Edge or SrcVertex.

Supported Platforms:: GPU

Examples

>>> import math
>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestAvg(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"x": x})
...         for v in g.dst_vertex:
...             v.h = g.avg([u.x for u in v.innbs])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestAvg()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> NAN = 1e9
>>> for row in ret:
...     if math.isnan(row[0]):
...         row[0] = NAN
>>> print(ret)
    [[1.0], [1.0], [1000000000.0], [0.0], [1.5], [2.0], [1.0], [1000000000.0], [1.0]]

dot(feat_x, feat_y)[source]

Dot mul operation for two node Tensors.

Parameters

feat_x (SrcVertex feature or DstVertex feature) – the arttribute of SrcVertex or DstVertex represent feature tensor of graph nodes with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.
feat_y (SrcVertex feature or DstVertex feature) – the arttribute of SrcVertex or DstVertex represent feature tensor of graph nodes with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.

Returns

Tensor, a tensor with shape \((N, 1)\), \(N\) is the number of nodes of the graph.

Raises

TypeError – If feat_x is not in the ‘mul’ operation support types [Tensor,Number,List,Tuple].
TypeError – If feat_y is not in the ‘mul’ operation support types [Tensor,Number,List,Tuple].

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestDot(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"src": x, "dst": x})
...         for v in g.dst_vertex:
...             v.h = [g.dot(v.src, u.dst) for u in v.innbs]
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestDot()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[2.0], [1.0], [2.0], [2.0], [0.0], [0.0], [2.0], [0.0], [1.0], [1.0], [1.0]]

property dst_idx: A tensor with shape \((N\_EDGES)\), represents the destination node index of COO edge matrix.

property dst_vertex

Return a list of destination vertex that only supports iterate its innbs.

Examples

>>> for v in g.dst_vertex:
...     pass

in_degree()[source]

Get the in degree of each node in a graph.

Returns: Tensor, a tensor with shape \((N, 1)\), represent the in degree of each node, \(N\) is the number of nodes of the graph.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
...
>>> class TestInDegree(GNNCell):
...     def construct(self, g: Graph):
...         return g.in_degree()
...
>>> ret = TestInDegree()(*graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1], [2], [0], [1], [2], [1], [1], [0], [3]]

max(neigh_feat)[source]

Aggregating node features from their neighbour and generates a node-level representation by aggregate function ‘max’.

Parameters: neigh_feat (List[SrcVertex feature or Edge feature]) – a list of SrcVertex or Edge attributes represents the neighbour nodes or edges feature, with shape \((N, F)\), \(N\) is the number of SrcVertex or Edge, \(F\) is the feature dimension of the SrcVertex or Edge attribute.
Returns: Tensor, a tensor with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.
Raises: TypeError – If neigh_feat is not a list of Edge or SrcVertex.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestMax(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"x": x})
...         for v in g.dst_vertex:
...             v.h = g.max([u.x for u in v.innbs])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestMax()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1.0], [1.0], [0.0], [0.0], [2.0], [2.0], [1.0], [0.0], [1.0]]

min(neigh_feat)[source]

Aggregating node features from their neighbour and generates a node-level representation by aggregate function ‘min’.

Parameters: neigh_feat (List[SrcVertex feature or Edge feature]) – a list of SrcVertex or Edge attributes represents the neighbour nodes or edges feature, with shape \((N, F)\), \(N\) is the number of SrcVertex or Edge, \(F\) is the feature dimension of the SrcVertex or Edge attribute.
Returns: Tensor, a tensor with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.
Raises: TypeError – If neigh_feat is not a list of Edge or SrcVertex.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestMin(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"x": x})
...         for v in g.dst_vertex:
...             v.h = g.min([u.x for u in v.innbs])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestMin()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]]

property n_edges: An integer, represent the edges count of the graph.

property n_nodes: An integer, represent the nodes count of the graph.

out_degree()[source]

Get the out degree of each node in a graph.

Returns: Tensor, a tensor with shape \((N, 1)\), represent the out degree of each node, \(N\) is the number of nodes of the graph.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
...
>>> class TestOutDegree(GNNCell):
...     def construct(self, g: Graph):
...         return g.out_degree()
...
>>> ret = TestOutDegree()(*graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1], [0], [2], [1], [1], [2], [1], [0], [3]]

set_dst_attr(feat_dict)[source]

Set attributes for destination vetices in vertex-centric environment Keys will be attribute’s name, values will be attributes’ data.

Parameters: feat_dict (Dict) – key type: str, value type: recommend tensor of shape \((N\_NODES, F)\), \(F\) is the dimension of the node feature.
Raises: TypeError – If feat_dict is not a Dict.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestSetDstAttr(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_dst_attr({"h": x})
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestSetDstAttr()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1.0], [2.0], [1.0], [2.0], [0.0], [1.0], [2.0], [3.0], [1.0]]

set_edge_attr(feat_dict)[source]

Set attributes for edges in vertex-centric environment. Keys will be attribute’s name, values will be attributes’ data.

Parameters: feat_dict (Dict) – key type: str, value type: recommend feature tensor of shape \((N\_EDGES, *)\), \(*\) is the shape of the feature per edge. Recommend the shape of value is \((N\_EDGES, 1)\) when the feature dimension is 1.
Raises: TypeError – If feat_dict is not a Dict.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
>>> edge_feat = ms.Tensor([[1], [2], [1], [3], [1], [4], [1], [5], [1], [1], [1]], ms.float32)
...
>>> class TestSetEdgeAttr(GNNCell):
...     def construct(self, nh, eh, g: Graph):
...         g.set_vertex_attr({"nh": nh})
...         g.set_edge_attr({"eh": eh})
...         for v in g.dst_vertex:
...             v.h = g.sum([u.nh * e.eh for u, e in v.inedges])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestSetEdgeAttr()(node_feat, edge_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[2.0], [2.0], [0.0], [0.0], [14.0], [6.0], [1.0], [0.0], [3.0]]

set_graph_attr(feat_dict)[source]

Set attributes for the whole graph in vertex-centric environment. Keys will be attribute’s name, values will be attributes’ data.

Parameters: feat_dict (Dict) – key type: str, value type: recommend feature tensor for the whole graph.
Raises: TypeError – If feat_dict is not a Dict.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> g_attr = ms.Tensor([[0.0, 1.0], [0.0, 1.0], [0.0, 1.0]], ms.float32)
>>> v_attr = ms.Tensor([1.0, 1.0], ms.float32)
...
>>> class TestSetGraphAttr(GNNCell):
...     def construct(self, vh, gh, g: Graph):
...         g.set_graph_attr({"x": gh})
...         g.set_vertex_attr({"h": vh})
...         for v in g.dst_vertex:
...             v.h = g.sum([u.h * g.x for u in v.innbs])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestSetGraphAttr()(v_attr, g_attr, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[0.0, 1.0], [0.0, 2.0], [0.0, 0.0], [0.0, 0.0],
     [0.0, 0.0], [0.0, 0.0], [0.0, 0.0], [0.0, 0.0], [0.0, 0.0]]

set_src_attr(feat_dict)[source]

Set attributes for source vertices in vertex-centric environment. Keys will be attribute’s name, values will be attributes’ data.

Parameters: feat_dict (Dict) – key type: str, value type: recommend tensor of shape \((N\_NODES, F)\), \(F\) is the dimension of the node feature.
Raises: TypeError – If feat_dict is not a Dict.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestSetSrcAttr(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_src_attr({"h": x})
...         return [u.h for u in g.src_vertex]
...
>>> ret = TestSetSrcAttr()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1.0], [2.0], [1.0], [2.0], [0.0], [1.0], [2.0], [3.0], [1.0]]

set_vertex_attr(feat_dict)[source]

Set attributes for vertices in vertex-centric environment. Keys will be attribute’s name, values will be attributes’ data.

Note

set_vertex_attr is equals to set_src_attr + set_dst_attr.

Parameters: feat_dict (Dict) – key type: str, value type: recommend tensor of shape \((N\_NODES, F)\), \(F\) is the dimension of the node feature.
Raises: TypeError – If feat_dict is not a Dict.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestSetVertexAttr(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"h": x})
...         return [v.h for v in g.dst_vertex] * [u.h for u in g.src_vertex]
...
>>> ret = TestSetVertexAttr()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1.0], [4.0], [1.0], [4.0], [0.0], [1.0], [4.0], [9.0], [1.0]]

property src_idx: A tensor with shape \((N\_EDGES)\), represents the source node index of COO edge matrix.

property src_vertex

Return a list of vertex that only supports iterate with its outnbs

Examples

>>> for u in g.src_vertex:
...     pass

sum(neigh_feat)[source]

Aggregating node features from their neighbour and generates a node-level representation by aggregate function ‘sum’.

Parameters: neigh_feat (List[SrcVertex feature or Edge feature]) – a list of SrcVertex or Edge attribute represents the neighbour nodes or edges feature, with shape \((N, F)\), \(N\) is the number of SrcVertex or Edge, \(F\) is the feature dimension of the SrcVertex or Edge attribute.
Returns: Tensor, a tensor with shape \((N, F)\), \(N\) is the number of nodes of the graph, \(F\) is the feature dimension of the node.
Raises: TypeError – If neigh_feat is not a list of Edge or SrcVertex.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import Graph, GraphField
>>> from mindspore_gl.nn import GNNCell
>>> n_nodes = 9
>>> n_edges = 11
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6, 8, 8, 8], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4, 8, 8, 8], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
>>> node_feat = ms.Tensor([[1], [2], [1], [2], [0], [1], [2], [3], [1]], ms.float32)
...
>>> class TestSum(GNNCell):
...     def construct(self, x, g: Graph):
...         g.set_vertex_attr({"x": x})
...         for v in g.dst_vertex:
...             v.h = g.sum([u.x for u in v.innbs])
...         return [v.h for v in g.dst_vertex]
...
>>> ret = TestSum()(node_feat, *graph_field.get_graph()).asnumpy().tolist()
>>> print(ret)
    [[1.0], [2.0], [0.0], [0.0], [3.0], [2.0], [1.0], [0.0], [3.0]]

topk_edges(node_feat, k, sortby=None)[source]

Return a graph-level representation by a graph-wise top-k on node features.

If sortby is set to None, the function would perform top-k on all dimensions independently.

Parameters

node_feat (Tensor) – A tensor represent the node feature, with shape \((N\_NODES, F)\). \(F\) is the dimension of the node feature.
k (int) – Represent how many nodes for top-k.
sortby (int) – Sort according to which feature. If is None, all features are sorted independently. Default is None.

Note

The value participated in the sort by axis (all value if sortby is None) should be greater than zero. Due to the reason that we create zero value for padding and they may cover the features.

Returns

topk_output (Tensor) - a tensor with shape \((B, K, F)\), where \(B\) is the batch size of the input graph. \(K\) is the input ‘k’, \(F\) is the feature size.
topk_indices (Tensor), - a tensor with shape \((B, K)\) if sortby is set to None), where \(B\) is the batch size of the input graph, \(F\) is the feature size.

Raises

TypeError – If node_feat is not a Tensor.
TypeError – If k is not an int.
ValueError – If sortby is not an int.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import BatchedGraph, BatchedGraphField
>>> from mindspore_gl.nn import GNNCell
>>> node_feat = ms.Tensor([
...     [1, 2, 3, 4],
...     [2, 4, 1, 3],
...     [1, 3, 2, 4],
...     [9, 7, 5, 8],
...     [8, 7, 6, 5],
...     [8, 6, 4, 6],
...     [1, 2, 1, 1],
... ], ms.float32)
...
>>> n_nodes = 7
>>> n_edges = 8
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
...
>>> class TestTopkEdges(GNNCell):
...     def construct(self, x, g: Graph):
...         return g.topk_edges(x, 2, 1)
...
>>> output, indices = TestTopkEdges()(node_feat, *graph_field.get_graph())
>>> output = output.asnumpy().tolist()
>>> indices = indices.asnumpy().tolist()
>>> print(output)
    [[9.0, 7.0, 5.0, 8.0], [8.0, 7.0, 6.0, 5.0]]
>>> print(indices)
    [3, 4]

topk_nodes(node_feat, k, sortby=None)[source]

Return a graph-level representation by a graph-wise top-k on node features.

If sortby is set to None, the function would perform top-k on all dimensions independently.

Parameters

node_feat (Tensor) – A tensor represent the node feature, with shape \((N\_NODES, F)\). \(F\) is the dimension of the node feature.
k (int) – Represent how many nodes for top-k.
sortby (int) – Sort according to which feature. If is None, all features are sorted independently. Default is None.

Note

The value participated in the sort by axis (all value if sortby is None) should be greater than zero. Due to the reason that we create zero value for padding and they may cover the features.

Returns

topk_output (Tensor) - a tensor with shape \((B, K, F)\), where \(B\) is the batch size of the input graph. \(K\) is the input ‘k’, \(F\) is the feature size.
topk_indices (Tensor), - a tensor with shape \((B, K)\) if sortby is set to None), where \(B\) is the batch size of the input graph, \(F\) is the feature size.

Raises

TypeError – If node_feat is not a Tensor.
TypeError – If k is not an int.
ValueError – If sortby is not an int.

Supported Platforms:: GPU

Examples

>>> import mindspore as ms
>>> from mindspore_gl import BatchedGraph, BatchedGraphField
>>> from mindspore_gl.nn import GNNCell
>>> node_feat = ms.Tensor([
...     [1, 2, 3, 4],
...     [2, 4, 1, 3],
...     [1, 3, 2, 4],
...     [9, 7, 5, 8],
...     [8, 7, 6, 5],
...     [8, 6, 4, 6],
...     [1, 2, 1, 1],
... ], ms.float32)
...
>>> n_nodes = 7
>>> n_edges = 8
>>> src_idx = ms.Tensor([0, 2, 2, 3, 4, 5, 5, 6], ms.int32)
>>> dst_idx = ms.Tensor([1, 0, 1, 5, 3, 4, 6, 4], ms.int32)
>>> graph_field = GraphField(src_idx, dst_idx, n_nodes, n_edges)
...
>>> class TestTopkNodes(GNNCell):
...     def construct(self, x, g: Graph):
...         return g.topk_nodes(x, 2, 1)
...
>>> output, indices = TestTopkNodes()(node_feat, *graph_field.get_graph())
>>> output = output.asnumpy().tolist()
>>> indices = indices.asnumpy().tolist()
>>> print(output)
    [[9.0, 7.0, 5.0, 8.0], [8.0, 7.0, 6.0, 5.0]]
>>> print(indices)
    [3, 4]