import torch import copy import dgl.function as fn class Graph: def __init__(self, graph, features, labels, edge_index, infmatrix, lap, hop): self.graph = graph self.features = features self.labels = labels self.edge_index = edge_index self.infmatrix = infmatrix self.lap = lap self.hop = hop self.embeddings = [] self.distances = [] D_invsqrt = torch.pow(self.graph.in_degrees().float().clamp(min=1), -0.5).unsqueeze(-1) self.xlx = torch.sigmoid(torch.diag(torch.spmm(torch.spmm(self.features.T, self.lap), self.features))) tempgraph = copy.deepcopy(self.graph) tempinfmatrix = torch.spmm(self.infmatrix, self.features) tempembedding = self.features self.distances.append(torch.abs(tempembedding - tempinfmatrix)) self.embeddings.append(tempembedding) with tempgraph.local_scope(): for i in range(1, hop): tempgraph.ndata['h'] = tempembedding * D_invsqrt tempgraph.update_all(fn.copy_u('h', 'm'), fn.sum('m', 'h')) tempembedding = tempgraph.ndata.pop('h') * D_invsqrt self.distances.append(torch.abs(tempembedding - tempinfmatrix)) self.embeddings.append(tempembedding)