model.py

import tensorflow as tf
from layers import GraphConvolution, GraphConvolutionSparse, InnerProductDecoder,DotProductDecoder,BilinearDecoder,MLP
from utils import *


class GCNModel():
    def __init__(self, placeholders, num_features, emb_dim, features_nonzero, adj_nonzero, num_r, name, act=tf.nn.elu):
        self.name = name
        self.inputs = placeholders['features']
        self.input_dim = num_features
        self.emb_dim = emb_dim
        self.features_nonzero = features_nonzero
        self.adj_nonzero = adj_nonzero
        self.adj = placeholders['adj']
        self.dropout = placeholders['dropout']
        self.adjdp = placeholders['adjdp']
        self.act = act
        self.att = tf.Variable(tf.constant([0.4, 0.25, 0.25,0.2]))
        self.num_r = num_r
        with tf.variable_scope(self.name):
            self.build()

    def build(self):
        self.adj = dropout_sparse(self.adj, 1-self.adjdp, self.adj_nonzero)
        self.hidden1 = GraphConvolutionSparse(
            name='gcn_sparse_layer',
            input_dim=self.input_dim,
            output_dim=self.emb_dim,
            adj=self.adj,
            features_nonzero=self.features_nonzero,
            dropout=self.dropout,
            act=self.act)(self.inputs)
        self.hidden2 = GraphConvolution(
            name='gcn_dense_layer',
            input_dim=self.emb_dim,
            output_dim=self.emb_dim,
            adj=self.adj,
            dropout=self.dropout,
            act=self.act)(self.hidden1)
        self.hidden3 = MLP(
            name='mlp_layer3',
            input_dim=self.emb_dim,
            hidden_dims=[self.emb_dim],
            output_dim=self.emb_dim,
            dropout=self.dropout,
            act=self.act)(self.hidden2)

        self.emb = GraphConvolution(
            name='gcn_dense_layer2',
            input_dim=self.emb_dim,
            output_dim=self.emb_dim,
            adj=self.adj,
            dropout=self.dropout,
            act=self.act)(self.hidden3)
        self.embeddings = self.hidden1 * \
            self.att[0]+self.hidden2*self.att[1]+ self.hidden3 * self.att[2]+self.emb*self.att[3]
        self.reconstructions = BilinearDecoder(input_dim=self.emb_dim, name='gcn_decoder',  num_r=self.num_r,rate=self.dropout, act=tf.nn.sigmoid)(self.embeddings)