import math
from typing import List, Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import Tensor
from greatx.nn.layers import GCNConv, Sequential, activations
from greatx.utils import wrapper
bce = F.binary_cross_entropy_with_logits
class Discriminator(nn.Module):
def __init__(self, in_channels: int):
super().__init__()
self.weight = nn.Parameter(Tensor(in_channels, in_channels))
@staticmethod
def uniform(size, tensor):
bound = 1.0 / math.sqrt(size)
if tensor is not None:
tensor.data.uniform_(-bound, bound)
def reset_parameters(self):
size = self.weight.size(0)
self.uniform(size, self.weight)
def forward(self, x: Tensor, summary: Tensor) -> Tensor:
""""""
x = torch.matmul(x, torch.matmul(self.weight, summary))
return x
[docs]class DGI(nn.Module):
r"""Deep Graph Infomax (DGI) from the
`"Deep Graph Infomax"
<https://arxiv.org/abs/1809.10341>`_ paper (ICLR'19)
Parameters
----------
in_channels : int,
the input dimensions of model
hids : List[int], optional
the number of hidden units for each hidden layer, by default [512]
acts : List[str], optional
the activation function for each hidden layer, by default ['prelu']
dropout : float, optional
the dropout ratio of model, by default 0.0
bias : bool, optional
whether to use bias in the layers, by default True
bn: bool, optional
whether to use :class:`BatchNorm1d` after the convolution layer,
by default False
Examples
--------
>>> # DGI with one hidden layer
>>> model = DGI(100)
>>> # DGI with two hidden layers
>>> model = DGI(100, hids=[32, 16], acts=['relu', 'elu'])
>>> # DGI with two hidden layers, without first activation
>>> model = DGI(100, hids=[32, 16], acts=[None, 'relu'])
>>> # DGI with deep architectures, each layer has elu activation
>>> model = DGI(100, hids=[16]*8, acts=['elu'])
Reference:
* Author's code: https://github.com/PetarV-/DGI
"""
@wrapper
def __init__(
self,
in_channels: int,
hids: List[int] = [512],
acts: List[str] = ['prelu'],
dropout: float = 0.,
bias: bool = True,
bn: bool = False,
):
super().__init__()
encoder = []
for hid, act in zip(hids, acts):
encoder.append(GCNConv(in_channels, hid, bias=bias))
if bn:
encoder.append(nn.BatchNorm1d(hid))
encoder.append(activations.get(act))
encoder.append(nn.Dropout(dropout))
in_channels = hid
self.encoder = Sequential(*encoder)
self.discriminator = Discriminator(in_channels)
self.reset_parameters()
[docs] @staticmethod
def corruption(x: Tensor) -> Tensor:
return x[torch.randperm(x.size(0))]
[docs] def reset_parameters(self):
self.encoder.reset_parameters()
self.discriminator.reset_parameters()
[docs] def encode(
self,
x: Tensor,
edge_index: Tensor,
edge_weight: Optional[Tensor] = None,
) -> Tensor:
z = self.encoder(x, edge_index, edge_weight)
return z
[docs] def forward(
self,
x: Tensor,
edge_index: Tensor,
edge_weight: Optional[Tensor] = None,
) -> Tensor:
""""""
z1 = self.encode(x, edge_index, edge_weight) # view1
z2 = self.encode(self.corruption(x), edge_index, edge_weight) # view2
summary = torch.sigmoid(z1.mean(dim=0)) # global
# global-local contrasting
pos = self.discriminator(z1, summary).squeeze()
# global-local contrasting
neg = self.discriminator(z2, summary).squeeze()
return pos, neg
[docs] def loss(self, postive: Tensor, negative: Tensor) -> Tensor:
loss = bce(postive, postive.new_ones(postive.size(0))) + \
bce(negative, negative.new_zeros(negative.size(0)))
return loss