Skip to content

GitLab

Unverified Commit baace710 authored by Zihao Ye's avatar Zihao Ye Committed by GitHub
Browse files

[documentation] Improve the docstring of global pooling layer (#2437) 

* upd

* fix-link

* lint
parent 2f28f791
Hide whitespace changes
Inline Side-by-side
Showing with 451 additions and 44 deletions
docs/source/api/python/nn.pytorch.rst
View file @ baace710
...@@ -190,6 +190,13 @@ SortPooling ...@@ -190,6 +190,13 @@ SortPooling
:members: :members:
:show-inheritance: :show-inheritance:
WeightAndSum
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: dgl.nn.pytorch.glob.WeightAndSum
:members:
:show-inheritance:
GlobalAttentionPooling GlobalAttentionPooling
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
......
python/dgl/nn/pytorch/glob.py
View file @ baace710
...@@ -37,20 +37,19 @@ class SumPooling(nn.Module): ...@@ -37,20 +37,19 @@ class SumPooling(nn.Module):
>>> import torch as th >>> import torch as th
>>> from dgl.nn.pytorch.glob import SumPooling >>> from dgl.nn.pytorch.glob import SumPooling
>>> >>>
>>> g1 = dgl.DGLGraph() >>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1.add_nodes(2) >>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>> g1_node_feats = th.ones(2,5)
>>> >>>
>>> g2 = dgl.DGLGraph() >>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2.add_nodes(3) >>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>> g2_node_feats = th.ones(3,5)
>>> >>>
>>> sumpool = SumPooling() >>> sumpool = SumPooling() # create a sum pooling layer
Case 1: Input a single graph Case 1: Input a single graph
>>> sumpool(g1, g1_node_feats) >>> sumpool(g1, g1_node_feats)
tensor([[2., 2., 2., 2., 2.]]) tensor([[10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,
10., 10.]])
Case 2: Input a batch of graphs Case 2: Input a batch of graphs
...@@ -60,8 +59,10 @@ class SumPooling(nn.Module): ...@@ -60,8 +59,10 @@ class SumPooling(nn.Module):
>>> batch_f = th.cat([g1_node_feats, g2_node_feats]) >>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>> >>>
>>> sumpool(batch_g, batch_f) >>> sumpool(batch_g, batch_f)
tensor([[2., 2., 2., 2., 2.], tensor([[10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10., 10.,
[3., 3., 3., 3., 3.]]) 10., 10.],
[20., 20., 20., 20., 20., 20., 20., 20., 20., 20., 20., 20., 20., 20.,
20., 20.]])
""" """
def __init__(self): def __init__(self):
super(SumPooling, self).__init__() super(SumPooling, self).__init__()
...@@ -115,20 +116,18 @@ class AvgPooling(nn.Module): ...@@ -115,20 +116,18 @@ class AvgPooling(nn.Module):
>>> import torch as th >>> import torch as th
>>> from dgl.nn.pytorch.glob import AvgPooling >>> from dgl.nn.pytorch.glob import AvgPooling
>>> >>>
>>> g1 = dgl.DGLGraph() >>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1.add_nodes(2) >>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>> g1_node_feats = th.ones(2,5)
>>> >>>
>>> g2 = dgl.DGLGraph() >>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2.add_nodes(3) >>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>> g2_node_feats = th.ones(3,5)
>>> >>>
>>> avgpool = AvgPooling() >>> avgpool = AvgPooling() # create an average pooling layer
Case 1: Input single graph Case 1: Input single graph
>>> avgpool(g1, g1_node_feats) >>> avgpool(g1, g1_node_feats)
tensor([[1., 1., 1., 1., 1.]]) tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
Case 2: Input a batch of graphs Case 2: Input a batch of graphs
...@@ -138,8 +137,8 @@ class AvgPooling(nn.Module): ...@@ -138,8 +137,8 @@ class AvgPooling(nn.Module):
>>> batch_f = th.cat([g1_node_feats, g2_node_feats]) >>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>> >>>
>>> avgpool(batch_g, batch_f) >>> avgpool(batch_g, batch_f)
tensor([[1., 1., 1., 1., 1.], tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1.]]) [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
""" """
def __init__(self): def __init__(self):
super(AvgPooling, self).__init__() super(AvgPooling, self).__init__()
...@@ -193,20 +192,18 @@ class MaxPooling(nn.Module): ...@@ -193,20 +192,18 @@ class MaxPooling(nn.Module):
>>> import torch as th >>> import torch as th
>>> from dgl.nn.pytorch.glob import MaxPooling >>> from dgl.nn.pytorch.glob import MaxPooling
>>> >>>
>>> g1 = dgl.DGLGraph() >>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1.add_nodes(2) >>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>> g1_node_feats = th.ones(2,5)
>>> >>>
>>> g2 = dgl.DGLGraph() >>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2.add_nodes(3) >>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>> g2_node_feats = th.ones(3,5)
>>> >>>
>>> maxpool = MaxPooling() >>> maxpool = MaxPooling() # create a max pooling layer
Case 1: Input a single graph Case 1: Input a single graph
>>> maxpool(g1, g1_node_feats) >>> maxpool(g1, g1_node_feats)
tensor([[1., 1., 1., 1., 1.]]) tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
Case 2: Input a batch of graphs Case 2: Input a batch of graphs
...@@ -216,8 +213,8 @@ class MaxPooling(nn.Module): ...@@ -216,8 +213,8 @@ class MaxPooling(nn.Module):
>>> batch_f = th.cat([g1_node_feats, g2_node_feats]) >>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>> >>>
>>> maxpool(batch_g, batch_f) >>> maxpool(batch_g, batch_f)
tensor([[1., 1., 1., 1., 1.], tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1.]]) [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
""" """
def __init__(self): def __init__(self):
super(MaxPooling, self).__init__() super(MaxPooling, self).__init__()
...@@ -271,20 +268,19 @@ class SortPooling(nn.Module): ...@@ -271,20 +268,19 @@ class SortPooling(nn.Module):
>>> import torch as th >>> import torch as th
>>> from dgl.nn.pytorch.glob import SortPooling >>> from dgl.nn.pytorch.glob import SortPooling
>>> >>>
>>> g1 = dgl.DGLGraph() >>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1.add_nodes(2) >>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>> g1_node_feats = th.ones(2,5)
>>> >>>
>>> g2 = dgl.DGLGraph() >>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2.add_nodes(3) >>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>> g2_node_feats = th.ones(3,5)
>>> >>>
>>> sortpool = SortPooling(k=2) >>> sortpool = SortPooling(k=2) # create a sort pooling layer
Case 1: Input a single graph Case 1: Input a single graph
>>> sortpool(g1, g1_node_feats) >>> sortpool(g1, g1_node_feats)
tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]]) tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
Case 2: Input a batch of graphs Case 2: Input a batch of graphs
...@@ -294,8 +290,10 @@ class SortPooling(nn.Module): ...@@ -294,8 +290,10 @@ class SortPooling(nn.Module):
>>> batch_f = th.cat([g1_node_feats, g2_node_feats]) >>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>> >>>
>>> sortpool(batch_g, batch_f) >>> sortpool(batch_g, batch_f)
tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1.], tensor([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
[1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]]) 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
""" """
def __init__(self, k): def __init__(self, k):
super(SortPooling, self).__init__() super(SortPooling, self).__init__()
...@@ -349,6 +347,50 @@ class GlobalAttentionPooling(nn.Module): ...@@ -349,6 +347,50 @@ class GlobalAttentionPooling(nn.Module):
feat_nn : torch.nn.Module, optional feat_nn : torch.nn.Module, optional
A neural network applied to each feature before combining them with attention A neural network applied to each feature before combining them with attention
scores. scores.
Examples
--------
The following example uses PyTorch backend.
>>> import dgl
>>> import torch as th
>>> from dgl.nn.pytorch.glob import GlobalAttentionPooling
>>>
>>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1_node_feats = th.ones(10, 16) # feature dimension is 16
>>>
>>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2_node_feats = th.ones(20, 16)
>>>
>>> gate_nn = th.nn.Linear(16, 1) # the gate layer that maps node feature to scalar
>>> gap = GlobalAttentionPooling(gate_nn) # create a Global Attention Pooling layer
Case 1: Input a single graph
>>> gap(g1, g1_node_feats)
tensor([[1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],
grad_fn=<SegmentReduceBackward>)
Case 2: Input a batch of graphs
Build a batch of DGL graphs and concatenate all graphs' node features into one tensor.
>>> batch_g = dgl.batch([g1, g2])
>>> batch_f = th.cat([g1_node_feats, g2_node_feats], 0)
>>>
>>> gap(batch_g, batch_f)
tensor([[1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
[1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],
grad_fn=<SegmentReduceBackward>)
Notes
-----
See our `GGNN example <https://github.com/dmlc/dgl/tree/master/examples/pytorch/ggnn>`_
on how to use GatedGraphConv and GlobalAttentionPooling layer to build a Graph Neural
Networks that can solve Soduku.
""" """
def __init__(self, gate_nn, feat_nn=None): def __init__(self, gate_nn, feat_nn=None):
super(GlobalAttentionPooling, self).__init__() super(GlobalAttentionPooling, self).__init__()
...@@ -416,6 +458,57 @@ class Set2Set(nn.Module): ...@@ -416,6 +458,57 @@ class Set2Set(nn.Module):
The number of iterations. The number of iterations.
n_layers : int n_layers : int
The number of recurrent layers. The number of recurrent layers.
Examples
--------
The following example uses PyTorch backend.
>>> import dgl
>>> import torch as th
>>> from dgl.nn.pytorch.glob import Set2Set
>>>
>>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>>
>>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>>
>>> s2s = Set2Set(16, 2, 1) # create a Set2Set layer(n_iters=2, n_layers=1)
Case 1: Input a single graph
>>> s2s(g1, g1_node_feats)
tensor([[ 0.0162, -0.0080, -0.0864, -0.0488, 0.0408, 0.1140, -0.1426, -0.0379,
-0.1094, 0.2622, 0.1352, 0.0011, 0.1970, -0.1837, -0.0377, -0.0360,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],
grad_fn=<CatBackward>)
Case 2: Input a batch of graphs
Build a batch of DGL graphs and concatenate all graphs' node features into one tensor.
>>> batch_g = dgl.batch([g1, g2])
>>> batch_f = th.cat([g1_node_feats, g2_node_feats], 0)
>>>
>>> s2s(batch_g, batch_f)
tensor([[ 0.0162, -0.0080, -0.0864, -0.0488, 0.0408, 0.1140, -0.1426, -0.0379,
-0.1094, 0.2622, 0.1352, 0.0011, 0.1970, -0.1837, -0.0377, -0.0360,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
[ 0.0162, -0.0080, -0.0864, -0.0488, 0.0408, 0.1140, -0.1426, -0.0379,
-0.1094, 0.2622, 0.1352, 0.0011, 0.1970, -0.1837, -0.0377, -0.0360,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000,
1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],
grad_fn=<CatBackward>)
Notes
-----
Set2Set is widely used in molecular property predictions, see
`dgl-lifesci's MPNN example <https://github.com/awslabs/dgl-lifesci/blob/
ecd95c905479ec048097777039cf9a19cfdcf223/python/dgllife/model/model_zoo/
mpnn_predictor.py>`__
on how to use DGL's Set2Set layer in graph property prediction applications.
""" """
def __init__(self, input_dim, n_iters, n_layers): def __init__(self, input_dim, n_iters, n_layers):
super(Set2Set, self).__init__() super(Set2Set, self).__init__()
...@@ -477,7 +570,27 @@ class Set2Set(nn.Module): ...@@ -477,7 +570,27 @@ class Set2Set(nn.Module):
class MultiHeadAttention(nn.Module): class MultiHeadAttention(nn.Module):
r"""Multi-Head Attention block, used in Transformer, Set Transformer and so on.""" r"""Multi-Head Attention block, used in Transformer, Set Transformer and so on.
Parameters
----------
d_model : int
The feature size (input and output) in Multi-Head Attention layer.
num_heads : int
The number of heads.
d_head : int
The hidden size per head.
d_ff : int
The inner hidden size in the Feed-Forward Neural Network.
dropouth : float
The dropout rate of each sublayer.
dropouta : float
The dropout rate of attention heads.
Notes
-----
This module was used in SetTransformer layer.
"""
def __init__(self, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.): def __init__(self, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.):
super(MultiHeadAttention, self).__init__() super(MultiHeadAttention, self).__init__()
self.d_model = d_model self.d_model = d_model
...@@ -566,7 +679,27 @@ class MultiHeadAttention(nn.Module): ...@@ -566,7 +679,27 @@ class MultiHeadAttention(nn.Module):
class SetAttentionBlock(nn.Module): class SetAttentionBlock(nn.Module):
r"""SAB block mentioned in Set-Transformer paper.""" r"""SAB block introduced in Set-Transformer paper.
Parameters
----------
d_model : int
The feature size (input and output) in Multi-Head Attention layer.
num_heads : int
The number of heads.
d_head : int
The hidden size per head.
d_ff : int
The inner hidden size in the Feed-Forward Neural Network.
dropouth : float
The dropout rate of each sublayer.
dropouta : float
The dropout rate of attention heads.
Notes
-----
This module was used in SetTransformer layer.
"""
def __init__(self, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.): def __init__(self, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.):
super(SetAttentionBlock, self).__init__() super(SetAttentionBlock, self).__init__()
self.mha = MultiHeadAttention(d_model, num_heads, d_head, d_ff, self.mha = MultiHeadAttention(d_model, num_heads, d_head, d_ff,
...@@ -587,7 +720,27 @@ class SetAttentionBlock(nn.Module): ...@@ -587,7 +720,27 @@ class SetAttentionBlock(nn.Module):
class InducedSetAttentionBlock(nn.Module): class InducedSetAttentionBlock(nn.Module):
r"""ISAB block mentioned in Set-Transformer paper.""" r"""ISAB block introduced in Set-Transformer paper.
Parameters
----------
d_model : int
The feature size (input and output) in Multi-Head Attention layer.
num_heads : int
The number of heads.
d_head : int
The hidden size per head.
d_ff : int
The inner hidden size in the Feed-Forward Neural Network.
dropouth : float
The dropout rate of each sublayer.
dropouta : float
The dropout rate of attention heads.
Notes
-----
This module was used in SetTransformer layer.
"""
def __init__(self, m, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.): def __init__(self, m, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.):
super(InducedSetAttentionBlock, self).__init__() super(InducedSetAttentionBlock, self).__init__()
self.m = m self.m = m
...@@ -634,7 +787,27 @@ class InducedSetAttentionBlock(nn.Module): ...@@ -634,7 +787,27 @@ class InducedSetAttentionBlock(nn.Module):
class PMALayer(nn.Module): class PMALayer(nn.Module):
r"""Pooling by Multihead Attention, used in the Decoder Module of Set Transformer.""" r"""Pooling by Multihead Attention, used as the Decoder Module in Set Transformer.
Parameters
----------
d_model : int
The feature size (input and output) in Multi-Head Attention layer.
num_heads : int
The number of heads.
d_head : int
The hidden size per head.
d_ff : int
The kernel size in FFN (Positionwise Feed-Forward Network) layer.
dropouth : float
The dropout rate of each sublayer.
dropouta : float
The dropout rate of attention heads.
Notes
-----
This module was used in SetTransformer layer.
"""
def __init__(self, k, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.): def __init__(self, k, d_model, num_heads, d_head, d_ff, dropouth=0., dropouta=0.):
super(PMALayer, self).__init__() super(PMALayer, self).__init__()
self.k = k self.k = k
...@@ -714,6 +887,125 @@ class SetTransformerEncoder(nn.Module): ...@@ -714,6 +887,125 @@ class SetTransformerEncoder(nn.Module):
The dropout rate of each sublayer. The dropout rate of each sublayer.
dropouta : float dropouta : float
The dropout rate of attention heads. The dropout rate of attention heads.
Examples
--------
>>> import dgl
>>> import torch as th
>>> from dgl.nn.pytorch.glob import SetTransformerEncoder
>>>
>>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>>
>>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>>
>>> set_trans_enc = SetTransformerEncoder(16, 4, 4, 64) # create a settrans encoder.
Case 1: Input a single graph
>>> set_trans_enc(g1, g1_node_feats)
tensor([[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921]],
grad_fn=<NativeLayerNormBackward>)
Case 2: Input a batch of graphs
Build a batch of DGL graphs and concatenate all graphs' node features into one tensor.
>>> batch_g = dgl.batch([g1, g2])
>>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>>
>>> set_trans_enc(batch_g, batch_f)
tensor([[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921],
[ 0.1910, -0.7224, -1.7265, 1.8916, 0.4857, 0.8494, -1.4954, -0.2460,
1.8127, -0.8369, -0.8274, 0.4267, -0.5167, 0.6989, 0.1073, -0.0921]],
grad_fn=<NativeLayerNormBackward>)
See Also
--------
SetTransformerDecoder
Notes
-----
SetTransformerEncoder is not a readout layer, the tensor it returned is nodewise
representation instead out graphwise representation, and the SetTransformerDecoder
would return a graph readout tensor.
""" """
def __init__(self, d_model, n_heads, d_head, d_ff, def __init__(self, d_model, n_heads, d_head, d_ff,
n_layers=1, block_type='sab', m=None, dropouth=0., dropouta=0.): n_layers=1, block_type='sab', m=None, dropouth=0., dropouta=0.):
...@@ -788,6 +1080,69 @@ class SetTransformerDecoder(nn.Module): ...@@ -788,6 +1080,69 @@ class SetTransformerDecoder(nn.Module):
Dropout rate of each sublayer. Dropout rate of each sublayer.
dropouta : float dropouta : float
Dropout rate of attention heads. Dropout rate of attention heads.
Examples
--------
>>> import dgl
>>> import torch as th
>>> from dgl.nn.pytorch.glob import SetTransformerDecoder
>>>
>>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>>
>>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>>
>>> set_trans_dec = SetTransformerDecoder(16, 4, 4, 64, 1, 5) # define the layer
Case 1: Input a single graph
>>> set_trans_dec(g1, g1_node_feats)
tensor([[ 0.4635, 0.0275, -0.2637, 0.7168, 1.7655, 0.7687, -0.0031, -0.0562,
0.8125, 1.3546, -1.1208, -0.0198, -0.6820, -1.7057, -2.0887, 0.0310,
0.7714, -0.6568, -0.3391, -0.3792, 1.3881, 1.0091, -0.1747, -1.5370,
1.3287, 1.7640, 0.6081, -0.4512, -0.5583, -1.3908, -1.2931, -0.0894,
2.0826, -0.3916, 0.9458, -0.0952, 0.6316, -1.0485, -0.1104, 0.1100,
1.6364, 1.0246, -0.2355, -0.9597, -1.2427, -0.4817, -1.5894, -0.2764,
0.9830, -0.2319, -0.3492, -0.7830, 1.4185, -0.1799, 0.2063, -0.7108,
1.1052, 2.3187, 0.5359, -0.2413, -0.5357, -1.3557, -1.4125, -0.7675,
-0.0231, -0.2948, -0.3586, 0.6925, 0.6982, 1.1432, -0.5939, -1.6942,
1.6847, 1.5113, -0.7235, 0.0262, -1.4526, -0.0706, -1.3626, 0.8179]],
grad_fn=<ViewBackward>)
Case 2: Input a batch of graphs
Build a batch of DGL graphs and concatenate all graphs' node features into one tensor.
>>> batch_g = dgl.batch([g1, g2])
>>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>>
>>> set_trans_dec(batch_g, batch_f)
tensor([[ 0.4635, 0.0275, -0.2637, 0.7168, 1.7655, 0.7687, -0.0031, -0.0562,
0.8125, 1.3546, -1.1208, -0.0198, -0.6820, -1.7057, -2.0887, 0.0310,
0.7714, -0.6568, -0.3391, -0.3792, 1.3881, 1.0091, -0.1747, -1.5370,
1.3287, 1.7640, 0.6081, -0.4512, -0.5583, -1.3908, -1.2931, -0.0894,
2.0826, -0.3916, 0.9458, -0.0952, 0.6316, -1.0485, -0.1104, 0.1100,
1.6364, 1.0246, -0.2355, -0.9597, -1.2427, -0.4817, -1.5894, -0.2764,
0.9830, -0.2319, -0.3492, -0.7830, 1.4185, -0.1799, 0.2063, -0.7108,
1.1052, 2.3187, 0.5359, -0.2413, -0.5357, -1.3557, -1.4125, -0.7675,
-0.0231, -0.2948, -0.3586, 0.6925, 0.6982, 1.1432, -0.5939, -1.6942,
1.6847, 1.5113, -0.7235, 0.0262, -1.4526, -0.0706, -1.3626, 0.8179],
[ 0.4635, 0.0275, -0.2637, 0.7168, 1.7655, 0.7687, -0.0031, -0.0562,
0.8125, 1.3546, -1.1208, -0.0198, -0.6820, -1.7057, -2.0887, 0.0310,
0.7714, -0.6568, -0.3391, -0.3792, 1.3881, 1.0091, -0.1747, -1.5370,
1.3287, 1.7640, 0.6081, -0.4512, -0.5583, -1.3908, -1.2931, -0.0894,
2.0826, -0.3916, 0.9458, -0.0952, 0.6316, -1.0485, -0.1104, 0.1100,
1.6364, 1.0246, -0.2355, -0.9597, -1.2427, -0.4817, -1.5894, -0.2764,
0.9830, -0.2319, -0.3492, -0.7830, 1.4185, -0.1799, 0.2063, -0.7108,
1.1052, 2.3187, 0.5359, -0.2413, -0.5357, -1.3557, -1.4125, -0.7675,
-0.0231, -0.2948, -0.3586, 0.6925, 0.6982, 1.1432, -0.5939, -1.6942,
1.6847, 1.5113, -0.7235, 0.0262, -1.4526, -0.0706, -1.3626, 0.8179]],
grad_fn=<ViewBackward>)
See Also
--------
SetTransformerEncoder
""" """
def __init__(self, d_model, num_heads, d_head, d_ff, n_layers, k, dropouth=0., dropouta=0.): def __init__(self, d_model, num_heads, d_head, d_ff, n_layers, k, dropouth=0., dropouta=0.):
super(SetTransformerDecoder, self).__init__() super(SetTransformerDecoder, self).__init__()
...@@ -837,6 +1192,51 @@ class WeightAndSum(nn.Module): ...@@ -837,6 +1192,51 @@ class WeightAndSum(nn.Module):
---------- ----------
in_feats : int in_feats : int
Input atom feature size Input atom feature size
Examples
--------
The following example uses PyTorch backend.
>>> import dgl
>>> import torch as th
>>> from dgl.nn.pytorch.glob import WeightAndSum
>>>
>>> g1 = dgl.rand_graph(10, 20) # g1 is a random graph with 10 nodes and 20 edges
>>> g1_node_feats = th.ones(10, 16) # feature size is 16
>>>
>>> g2 = dgl.rand_graph(20, 50) # g2 is a random graph with 20 nodes and 50 edges
>>> g2_node_feats = th.ones(20, 16) # feature size is 16
>>>
>>> weight_and_sum = WeightAndSum(16) # create a weight and sum layer(in_feats=16)
Case 1: Input a single graph
>>> weight_and_sum(g1, g1_node_feats)
tensor([[5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436,
5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436]],
grad_fn=<SegmentReduceBackward>)
Case 2: Input a batch of graphs
Build a batch of DGL graphs and concatenate all graphs' node features into one tensor.
>>> batch_g = dgl.batch([g1, g2])
>>> batch_f = th.cat([g1_node_feats, g2_node_feats])
>>>
>>> sumpool(batch_g, batch_f)
tensor([[ 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436,
5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436, 5.1436],
[10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872,
10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872, 10.2872]],
grad_fn=<SegmentReduceBackward>)
Notes
-----
WeightAndSum module was commonly used in molecular property prediction networks,
see the GCN predictor in `dgl-lifesci <https://github.com/awslabs/dgl-lifesci/blob/
ae0491431804611ba466ff413f69d435789dbfd5/python/dgllife/model/model_zoo/
gcn_predictor.py>`__
to understand how to use WeightAndSum layer to get the graph readout output.
""" """
def __init__(self, in_feats): def __init__(self, in_feats):
super(WeightAndSum, self).__init__() super(WeightAndSum, self).__init__()
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment