layers.py

import torch
import torch.nn as nn
import numpy as np
import torch.nn.functional as F
import matplotlib.pyplot as plt
from initializations import *
from preprocessing import normalize_adj_torch


class GSRLayer(nn.Module):

    def __init__(self, hr_dim):
        super(GSRLayer, self).__init__()

        self.weights = torch.from_numpy(
            weight_variable_glorot(hr_dim)).type(torch.FloatTensor)
        self.weights = torch.nn.Parameter(
            data=self.weights, requires_grad=True)

    def forward(self, A, X):
        with torch.autograd.set_detect_anomaly(True):

            lr = A
            lr_dim = lr.shape[0]
            f = X
            eig_val_lr, U_lr = torch.symeig(lr, eigenvectors=True, upper=True)

            # U_lr = torch.abs(U_lr)
            eye_mat = torch.eye(lr_dim).type(torch.FloatTensor)
            s_d = torch.cat((eye_mat, eye_mat), 0)

            a = torch.matmul(self.weights, s_d)
            b = torch.matmul(a, torch.t(U_lr))
            f_d = torch.matmul(b, f)
            f_d = torch.abs(f_d)
            f_d = f_d.fill_diagonal_(1)
            adj = f_d

            X = torch.mm(adj, adj.t())
            X = (X + X.t())/2
            X = X.fill_diagonal_(1)
        return adj, torch.abs(X)


class GraphConvolution(nn.Module):
    """
    Simple GCN layer, similar to https://arxiv.org/abs/1609.02907
    """

    def __init__(self, in_features, out_features, dropout, act=F.relu):
        super(GraphConvolution, self).__init__()
        self.in_features = in_features
        self.out_features = out_features
        self.dropout = dropout
        self.act = act
        self.weight = torch.nn.Parameter(
            torch.FloatTensor(in_features, out_features))
        self.reset_parameters()

    def reset_parameters(self):
        torch.nn.init.xavier_uniform_(self.weight)

    def forward(self, input, adj):
        input = F.dropout(input, self.dropout, self.training)
        support = torch.mm(input, self.weight)
        output = torch.mm(adj, support)
        output = self.act(output)
        return output