code
stringlengths
17
6.64M
def train(args, adj_list, features, train_data, train_label, test_data, test_label, paras): with tf.Session() as sess: adj_data = adj_list net = GAS(session=sess, nodes=paras[0], class_size=paras[4], embedding_r=paras[1], embedding_u=paras[2], embedding_i=paras[3], h_u_size=paras[6], h_i_size=paras[7], encoding1=args.encoding1, encoding2=args.encoding2, encoding3=args.encoding3, encoding4=args.encoding4, gcn_dim=args.gcn_dim) sess.run(tf.global_variables_initializer()) t_start = time.clock() for epoch in range(args.epoch_num): train_loss = 0 train_acc = 0 count = 0 for index in range(0, paras[3], args.batch_size): (batch_data, batch_label) = get_data(index, args.batch_size, paras[3]) (loss, acc, pred, prob) = net.train(features, adj_data, batch_label, batch_data, args.learning_rate, args.momentum) print('batch loss: {:.4f}, batch acc: {:.4f}'.format(loss, acc)) train_loss += loss train_acc += acc count += 1 train_loss = (train_loss / count) train_acc = (train_acc / count) print('epoch{:d} : train_loss: {:.4f}, train_acc: {:.4f}'.format(epoch, train_loss, train_acc)) t_end = time.clock() print('train time=', '{:.5f}'.format((t_end - t_start))) print('Train end!') (test_acc, test_pred, test_probabilities, test_tags) = net.test(features, adj_data, test_label, test_data) print('test acc:', test_acc)
class GEM(Algorithm): def __init__(self, session, nodes, class_size, meta, embedding, encoding, hop): self.nodes = nodes self.meta = meta self.class_size = class_size self.embedding = embedding self.encoding = encoding self.hop = hop self.placeholders = {'a': tf.placeholder(tf.float32, [self.meta, self.nodes, self.nodes], 'adj'), 'x': tf.placeholder(tf.float32, [self.nodes, self.embedding], 'nxf'), 'batch_index': tf.placeholder(tf.int32, [None], 'index'), 't': tf.placeholder(tf.float32, [None, self.class_size], 'labels'), 'lr': tf.placeholder(tf.float32, [], 'learning_rate'), 'mom': tf.placeholder(tf.float32, [], 'momentum'), 'num_features_nonzero': tf.placeholder(tf.int32)} (loss, probabilities) = self.forward_propagation() (self.loss, self.probabilities) = (loss, probabilities) self.l2 = tf.contrib.layers.apply_regularization(tf.contrib.layers.l2_regularizer(0.01), tf.trainable_variables()) x = tf.ones_like(self.probabilities) y = tf.zeros_like(self.probabilities) self.pred = tf.where((self.probabilities > 0.5), x=x, y=y) print(self.pred.shape) self.correct_prediction = tf.equal(self.pred, self.placeholders['t']) self.accuracy = tf.reduce_mean(tf.cast(self.correct_prediction, 'float')) print('Forward propagation finished.') self.sess = session self.optimizer = tf.train.AdamOptimizer(self.placeholders['lr']) gradients = self.optimizer.compute_gradients((self.loss + self.l2)) capped_gradients = [(tf.clip_by_value(grad, (- 5.0), 5.0), var) for (grad, var) in gradients if (grad is not None)] self.train_op = self.optimizer.apply_gradients(capped_gradients) self.init = tf.global_variables_initializer() print('Backward propagation finished.') def forward_propagation(self): with tf.variable_scope('gem_embedding'): h = tf.get_variable(name='init_embedding', shape=[self.nodes, self.encoding], initializer=tf.contrib.layers.xavier_initializer()) for i in range(0, self.hop): f = GEMLayer(self.placeholders, self.nodes, self.meta, self.embedding, self.encoding) gem_out = f(inputs=h) h = tf.reshape(gem_out, [self.nodes, self.encoding]) print('GEM embedding over!') with tf.variable_scope('classification'): batch_data = tf.matmul(tf.one_hot(self.placeholders['batch_index'], self.nodes), h) W = tf.get_variable(name='weights', shape=[self.encoding, self.class_size], initializer=tf.contrib.layers.xavier_initializer()) b = tf.get_variable(name='bias', shape=[1, self.class_size], initializer=tf.zeros_initializer()) tf.transpose(batch_data, perm=[0, 1]) logits = (tf.matmul(batch_data, W) + b) u = tf.get_variable(name='u', shape=[1, self.encoding], initializer=tf.contrib.layers.xavier_initializer()) loss = tf.losses.sigmoid_cross_entropy(multi_class_labels=self.placeholders['t'], logits=logits) return (loss, tf.nn.sigmoid(logits)) def train(self, x, a, t, b, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict(x, a, t, b, learning_rate, momentum, self.placeholders) outs = self.sess.run([self.train_op, self.loss, self.accuracy, self.pred, self.probabilities], feed_dict=feed_dict) loss = outs[1] acc = outs[2] pred = outs[3] prob = outs[4] return (loss, acc, pred, prob) def test(self, x, a, t, b, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict(x, a, t, b, learning_rate, momentum, self.placeholders) (acc, pred, probabilities, tags) = self.sess.run([self.accuracy, self.pred, self.probabilities, self.correct_prediction], feed_dict=feed_dict) return (acc, pred, probabilities, tags)
def arg_parser(): parser = argparse.ArgumentParser() parser.add_argument('--seed', type=int, default=123, help='Random seed.') parser.add_argument('--dataset_str', type=str, default='dblp', help="['dblp','example']") parser.add_argument('--epoch_num', type=int, default=30, help='Number of epochs to train.') parser.add_argument('--batch_size', type=int, default=1000) parser.add_argument('--momentum', type=int, default=0.9) parser.add_argument('--learning_rate', default=0.001, help='the ratio of training set in whole dataset.') parser.add_argument('--hop', default=1, help='hop number') parser.add_argument('--k', default=16, help='gem layer unit') args = parser.parse_args() return args
def set_env(args): tf.reset_default_graph() np.random.seed(args.seed) tf.set_random_seed(args.seed)
def get_data(ix, int_batch, train_size): if ((ix + int_batch) >= train_size): ix = (train_size - int_batch) end = train_size else: end = (ix + int_batch) return (train_data[ix:end], train_label[ix:end])
def load_data(args): if (args.dataset_str == 'dblp'): (adj_list, features, train_data, train_label, test_data, test_label) = load_data_dblp() if (args.dataset_str == 'example'): (adj_list, features, train_data, train_label, test_data, test_label) = load_example_gem() node_size = features.shape[0] node_embedding = features.shape[1] class_size = train_label.shape[1] train_size = len(train_data) paras = [node_size, node_embedding, class_size, train_size] return (adj_list, features, train_data, train_label, test_data, test_label, paras)
def train(args, adj_list, features, train_data, train_label, test_data, test_label, paras): with tf.Session() as sess: adj_data = adj_list meta_size = len(adj_list) net = GEM(session=sess, class_size=paras[2], encoding=args.k, meta=meta_size, nodes=paras[0], embedding=paras[1], hop=args.hop) sess.run(tf.global_variables_initializer()) t_start = time.clock() for epoch in range(args.epoch_num): train_loss = 0 train_acc = 0 count = 0 for index in range(0, paras[3], args.batch_size): (batch_data, batch_label) = get_data(index, args.batch_size, paras[3]) (loss, acc, pred, prob) = net.train(features, adj_data, batch_label, batch_data, args.learning_rate, args.momentum) print('batch loss: {:.4f}, batch acc: {:.4f}'.format(loss, acc)) train_loss += loss train_acc += acc count += 1 train_loss = (train_loss / count) train_acc = (train_acc / count) print('epoch{:d} : train_loss: {:.4f}, train_acc: {:.4f}'.format(epoch, train_loss, train_acc)) t_end = time.clock() print('train time=', '{:.5f}'.format((t_end - t_start))) print('Train end!') (test_acc, test_pred, test_probabilities, test_tags) = net.test(features, adj_data, test_label, test_data) print('test acc:', test_acc)
def arg_parser(): parser = argparse.ArgumentParser() parser.add_argument('--seed', type=int, default=123, help='Random seed.') parser.add_argument('--dataset_str', type=str, default='dblp', help="['dblp','example']") parser.add_argument('--epoch_num', type=int, default=30, help='Number of epochs to train.') parser.add_argument('--batch_size', type=int, default=1000) parser.add_argument('--momentum', type=int, default=0.9) parser.add_argument('--learning_rate', default=0.001, help='the ratio of training set in whole dataset.') parser.add_argument('--dim', default=128) parser.add_argument('--lstm_hidden', default=128, help='lstm_hidden unit') parser.add_argument('--heads', default=1, help='gat heads') parser.add_argument('--layer_num', default=4, help='geniePath layer num') args = parser.parse_args() return args
def set_env(args): tf.reset_default_graph() np.random.seed(args.seed) tf.set_random_seed(args.seed)
def get_data(ix, int_batch, train_size): if ((ix + int_batch) >= train_size): ix = (train_size - int_batch) end = train_size else: end = (ix + int_batch) return (train_data[ix:end], train_label[ix:end])
def load_data(args): if (args.dataset_str == 'dblp'): (adj_list, features, train_data, train_label, test_data, test_label) = load_data_dblp() node_size = features.shape[0] node_embedding = features.shape[1] class_size = train_label.shape[1] train_size = len(train_data) paras = [node_size, node_embedding, class_size, train_size] return (adj_list, features, train_data, train_label, test_data, test_label, paras)
def train(args, adj_list, features, train_data, train_label, test_data, test_label, paras): with tf.Session() as sess: adj_data = adj_list net = GeniePath(session=sess, out_dim=paras[2], dim=args.dim, lstm_hidden=args.lstm_hidden, nodes=paras[0], in_dim=paras[1], heads=args.heads, layer_num=args.layer_num, class_size=paras[2]) sess.run(tf.global_variables_initializer()) t_start = time.clock() for epoch in range(args.epoch_num): train_loss = 0 train_acc = 0 count = 0 for index in range(0, paras[3], args.batch_size): (batch_data, batch_label) = get_data(index, args.batch_size, paras[3]) (loss, acc, pred, prob) = net.train(features, adj_data, batch_label, batch_data, args.learning_rate, args.momentum) print('batch loss: {:.4f}, batch acc: {:.4f}'.format(loss, acc)) train_loss += loss train_acc += acc count += 1 train_loss = (train_loss / count) train_acc = (train_acc / count) print('epoch{:d} : train_loss: {:.4f}, train_acc: {:.4f}'.format(epoch, train_loss, train_acc)) t_end = time.clock() print('train time=', '{:.5f}'.format((t_end - t_start))) print('Train end!') (test_acc, test_pred, test_probabilities, test_tags) = net.test(features, adj_data, test_label, test_data) print('test acc:', test_acc)
class MeanAggregator(Layer): '\n Aggregates via mean followed by matmul and non-linearity.\n ' def __init__(self, input_dim, output_dim, neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MeanAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([neigh_input_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_vecs = tf.nn.dropout(neigh_vecs, (1 - self.dropout)) self_vecs = tf.nn.dropout(self_vecs, (1 - self.dropout)) neigh_means = tf.reduce_mean(neigh_vecs, axis=1) from_neighs = tf.matmul(neigh_means, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class HeteMeanAggregator(Layer): '\n Aggregates via mean followed by matmul and non-linearity.\n ' def __init__(self, input_dim, output_dim, neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MeanAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([neigh_input_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_vecs = tf.nn.dropout(neigh_vecs, (1 - self.dropout)) self_vecs = tf.nn.dropout(self_vecs, (1 - self.dropout)) neigh_means = tf.reduce_mean(neigh_vecs, axis=1) from_neighs = tf.matmul(neigh_means, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class GCNAggregator(Layer): '\n Aggregates via mean followed by matmul and non-linearity.\n Same matmul parameters are used self vector and neighbor vectors.\n ' def __init__(self, input_dim, output_dim, neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(GCNAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['weights'] = glorot([neigh_input_dim, output_dim], name='neigh_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_vecs = tf.nn.dropout(neigh_vecs, (1 - self.dropout)) self_vecs = tf.nn.dropout(self_vecs, (1 - self.dropout)) means = tf.reduce_mean(tf.concat([neigh_vecs, tf.expand_dims(self_vecs, axis=1)], axis=1), axis=1) output = tf.matmul(means, self.vars['weights']) if self.bias: output += self.vars['bias'] return self.act(output)
class MaxPoolingAggregator(Layer): ' Aggregates via max-pooling over MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MaxPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 512 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 1024 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim)) neigh_h = tf.reduce_max(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class MeanPoolingAggregator(Layer): ' Aggregates via mean-pooling over MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MeanPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 512 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 1024 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim)) neigh_h = tf.reduce_mean(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class TwoMaxLayerPoolingAggregator(Layer): ' Aggregates via pooling over two MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(TwoMaxLayerPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim_1 = self.hidden_dim_1 = 512 hidden_dim_2 = self.hidden_dim_2 = 256 elif (model_size == 'big'): hidden_dim_1 = self.hidden_dim_1 = 1024 hidden_dim_2 = self.hidden_dim_2 = 512 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim_1, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) self.mlp_layers.append(Dense(input_dim=hidden_dim_1, output_dim=hidden_dim_2, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim_2, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim_2)) neigh_h = tf.reduce_max(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class SeqAggregator(Layer): ' Aggregates via a standard LSTM.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(SeqAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 128 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 256 with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim self.cell = tf.contrib.rnn.BasicLSTMCell(self.hidden_dim) def _call(self, inputs): (self_vecs, neigh_vecs) = inputs dims = tf.shape(neigh_vecs) batch_size = dims[0] initial_state = self.cell.zero_state(batch_size, tf.float32) used = tf.sign(tf.reduce_max(tf.abs(neigh_vecs), axis=2)) length = tf.reduce_sum(used, axis=1) length = tf.maximum(length, tf.constant(1.0)) length = tf.cast(length, tf.int32) with tf.variable_scope(self.name) as scope: try: (rnn_outputs, rnn_states) = tf.nn.dynamic_rnn(self.cell, neigh_vecs, initial_state=initial_state, dtype=tf.float32, time_major=False, sequence_length=length) except ValueError: scope.reuse_variables() (rnn_outputs, rnn_states) = tf.nn.dynamic_rnn(self.cell, neigh_vecs, initial_state=initial_state, dtype=tf.float32, time_major=False, sequence_length=length) batch_size = tf.shape(rnn_outputs)[0] max_len = tf.shape(rnn_outputs)[1] out_size = int(rnn_outputs.get_shape()[2]) index = ((tf.range(0, batch_size) * max_len) + (length - 1)) flat = tf.reshape(rnn_outputs, [(- 1), out_size]) neigh_h = tf.gather(flat, index) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) output = tf.add_n([from_self, from_neighs]) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
def uniform(shape, scale=0.05, name=None): 'Uniform init.' initial = tf.random_uniform(shape, minval=(- scale), maxval=scale, dtype=tf.float32) return tf.Variable(initial, name=name)
def glorot(shape, name=None): 'Glorot & Bengio (AISTATS 2010) init.' init_range = np.sqrt((6.0 / (shape[0] + shape[1]))) initial = tf.random_uniform(shape, minval=(- init_range), maxval=init_range, dtype=tf.float32) return tf.Variable(initial, name=name)
def zeros(shape, name=None): 'All zeros.' initial = tf.zeros(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def ones(shape, name=None): 'All ones.' initial = tf.ones(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def get_layer_uid(layer_name=''): 'Helper function, assigns unique layer IDs.' if (layer_name not in _LAYER_UIDS): _LAYER_UIDS[layer_name] = 1 return 1 else: _LAYER_UIDS[layer_name] += 1 return _LAYER_UIDS[layer_name]
class Layer(object): 'Base layer class. Defines basic API for all layer objects.\n Implementation inspired by keras (http://keras.io).\n # Properties\n name: String, defines the variable scope of the layer.\n logging: Boolean, switches Tensorflow histogram logging on/off\n\n # Methods\n _call(inputs): Defines computation graph of layer\n (i.e. takes input, returns output)\n __call__(inputs): Wrapper for _call()\n _log_vars(): Log all variables\n ' def __init__(self, **kwargs): allowed_kwargs = {'name', 'logging', 'model_size'} for kwarg in kwargs.keys(): assert (kwarg in allowed_kwargs), ('Invalid keyword argument: ' + kwarg) name = kwargs.get('name') if (not name): layer = self.__class__.__name__.lower() name = ((layer + '_') + str(get_layer_uid(layer))) self.name = name self.vars = {} logging = kwargs.get('logging', False) self.logging = logging self.sparse_inputs = False def _call(self, inputs): return inputs def __call__(self, inputs): with tf.name_scope(self.name): if (self.logging and (not self.sparse_inputs)): tf.summary.histogram((self.name + '/inputs'), inputs) outputs = self._call(inputs) if self.logging: tf.summary.histogram((self.name + '/outputs'), outputs) return outputs def _log_vars(self): for var in self.vars: tf.summary.histogram(((self.name + '/vars/') + var), self.vars[var])
class Dense(Layer): 'Dense layer.' def __init__(self, input_dim, output_dim, dropout=0.0, act=tf.nn.relu, placeholders=None, bias=True, featureless=False, sparse_inputs=False, **kwargs): super(Dense, self).__init__(**kwargs) self.dropout = dropout self.act = act self.featureless = featureless self.bias = bias self.input_dim = input_dim self.output_dim = output_dim self.sparse_inputs = sparse_inputs if sparse_inputs: self.num_features_nonzero = placeholders['num_features_nonzero'] with tf.variable_scope((self.name + '_vars')): self.vars['weights'] = tf.get_variable('weights', shape=(input_dim, output_dim), dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer(), regularizer=tf.contrib.layers.l2_regularizer(FLAGS.weight_decay)) if self.bias: self.vars['bias'] = zeros([output_dim], name='bias') if self.logging: self._log_vars() def _call(self, inputs): x = inputs x = tf.nn.dropout(x, (1 - self.dropout)) output = tf.matmul(x, self.vars['weights']) if self.bias: output += self.vars['bias'] return self.act(output)
def masked_logit_cross_entropy(preds, labels, mask): 'Logit cross-entropy loss with masking.' loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=preds, labels=labels) loss = tf.reduce_sum(loss, axis=1) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.maximum(tf.reduce_sum(mask), tf.constant([1.0])) loss *= mask return tf.reduce_mean(loss)
def masked_softmax_cross_entropy(preds, labels, mask): 'Softmax cross-entropy loss with masking.' loss = tf.nn.softmax_cross_entropy_with_logits(logits=preds, labels=labels) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.maximum(tf.reduce_sum(mask), tf.constant([1.0])) loss *= mask return tf.reduce_mean(loss)
def masked_l2(preds, actuals, mask): 'L2 loss with masking.' loss = tf.nn.l2(preds, actuals) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.reduce_mean(mask) loss *= mask return tf.reduce_mean(loss)
def masked_accuracy(preds, labels, mask): 'Accuracy with masking.' correct_prediction = tf.equal(tf.argmax(preds, 1), tf.argmax(labels, 1)) accuracy_all = tf.cast(correct_prediction, tf.float32) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.reduce_mean(mask) accuracy_all *= mask return tf.reduce_mean(accuracy_all)
class UniformNeighborSampler(Layer): '\n Uniformly samples neighbors.\n Assumes that adj lists are padded with random re-sampling\n ' def __init__(self, adj_info, **kwargs): super(UniformNeighborSampler, self).__init__(**kwargs) self.adj_info = adj_info def _call(self, inputs): (ids, num_samples) = inputs adj_lists = tf.nn.embedding_lookup(self.adj_info, ids) adj_lists = tf.transpose(tf.random_shuffle(tf.transpose(adj_lists))) adj_lists = tf.slice(adj_lists, [0, 0], [(- 1), num_samples]) return adj_lists
class DistanceNeighborSampler(Layer): '\n Sampling neighbors based on the feature consistency.\n ' def __init__(self, adj_info, **kwargs): super(DistanceNeighborSampler, self).__init__(**kwargs) self.adj_info = adj_info self.num_neighs = adj_info.shape[(- 1)] def _call(self, inputs): eps = 0.001 (ids, num_samples, features, batch_size) = inputs adj_lists = tf.gather(self.adj_info, ids) node_features = tf.gather(features, ids) feature_size = tf.shape(features)[(- 1)] node_feature_repeat = tf.tile(node_features, [1, self.num_neighs]) node_feature_repeat = tf.reshape(node_feature_repeat, [batch_size, self.num_neighs, feature_size]) neighbor_feature = tf.gather(features, adj_lists) distance = tf.sqrt(tf.reduce_sum(tf.square((node_feature_repeat - neighbor_feature)), (- 1))) prob = tf.exp((- distance)) prob_sum = tf.reduce_sum(prob, (- 1), keepdims=True) prob_sum = tf.tile(prob_sum, [1, self.num_neighs]) prob = tf.divide(prob, prob_sum) prob = tf.where((prob > eps), prob, (0 * prob)) samples_idx = tf.random.categorical(tf.math.log(prob), num_samples) selected = tf.batch_gather(adj_lists, samples_idx) return selected
class BipartiteEdgePredLayer(Layer): def __init__(self, input_dim1, input_dim2, placeholders, dropout=False, act=tf.nn.sigmoid, loss_fn='xent', neg_sample_weights=1.0, bias=False, bilinear_weights=False, **kwargs): '\n Basic class that applies skip-gram-like loss\n (i.e., dot product of node+target and node and negative samples)\n Args:\n bilinear_weights: use a bilinear weight for affinity calculation: u^T A v. If set to\n false, it is assumed that input dimensions are the same and the affinity will be \n based on dot product.\n ' super(BipartiteEdgePredLayer, self).__init__(**kwargs) self.input_dim1 = input_dim1 self.input_dim2 = input_dim2 self.act = act self.bias = bias self.eps = 1e-07 self.margin = 0.1 self.neg_sample_weights = neg_sample_weights self.bilinear_weights = bilinear_weights if dropout: self.dropout = placeholders['dropout'] else: self.dropout = 0.0 self.output_dim = 1 with tf.variable_scope((self.name + '_vars')): if bilinear_weights: self.vars['weights'] = tf.get_variable('pred_weights', shape=(input_dim1, input_dim2), dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer()) if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if (loss_fn == 'xent'): self.loss_fn = self._xent_loss elif (loss_fn == 'skipgram'): self.loss_fn = self._skipgram_loss elif (loss_fn == 'hinge'): self.loss_fn = self._hinge_loss if self.logging: self._log_vars() def affinity(self, inputs1, inputs2): ' Affinity score between batch of inputs1 and inputs2.\n Args:\n inputs1: tensor of shape [batch_size x feature_size].\n ' if self.bilinear_weights: prod = tf.matmul(inputs2, tf.transpose(self.vars['weights'])) self.prod = prod result = tf.reduce_sum((inputs1 * prod), axis=1) else: result = tf.reduce_sum((inputs1 * inputs2), axis=1) return result def neg_cost(self, inputs1, neg_samples, hard_neg_samples=None): ' For each input in batch, compute the sum of its affinity to negative samples.\n\n Returns:\n Tensor of shape [batch_size x num_neg_samples]. For each node, a list of affinities to\n negative samples is computed.\n ' if self.bilinear_weights: inputs1 = tf.matmul(inputs1, self.vars['weights']) neg_aff = tf.matmul(inputs1, tf.transpose(neg_samples)) return neg_aff def loss(self, inputs1, inputs2, neg_samples): ' negative sampling loss.\n Args:\n neg_samples: tensor of shape [num_neg_samples x input_dim2]. Negative samples for all\n inputs in batch inputs1.\n ' return self.loss_fn(inputs1, inputs2, neg_samples) def _xent_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) true_xent = tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(aff), logits=aff) negative_xent = tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.zeros_like(neg_aff), logits=neg_aff) loss = (tf.reduce_sum(true_xent) + (self.neg_sample_weights * tf.reduce_sum(negative_xent))) return loss def _skipgram_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) neg_cost = tf.log(tf.reduce_sum(tf.exp(neg_aff), axis=1)) loss = tf.reduce_sum((aff - neg_cost)) return loss def _hinge_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) diff = tf.nn.relu(tf.subtract(neg_aff, (tf.expand_dims(aff, 1) - self.margin)), name='diff') loss = tf.reduce_sum(diff) self.neg_shape = tf.shape(neg_aff) return loss def weights_norm(self): return tf.nn.l2_norm(self.vars['weights'])
class SupervisedGraphconsis(models.SampleAndAggregate): 'Implementation of supervised GraphConsis.' def __init__(self, num_classes, placeholders, features, adj, degrees, layer_infos, concat=True, aggregator_type='mean', model_size='small', sigmoid_loss=False, identity_dim=0, num_re=3, **kwargs): '\n Args:\n - placeholders: Stanford TensorFlow placeholder object.\n - features: Numpy array with node features.\n - adj: Numpy array with adjacency lists (padded with random re-samples)\n - degrees: Numpy array with node degrees. \n - layer_infos: List of SAGEInfo namedtuples that describe the parameters of all \n the recursive layers. See SAGEInfo definition above. It contains *numer_re* lists of layer_info\n - concat: whether to concatenate during recursive iterations\n - aggregator_type: how to aggregate neighbor information\n - model_size: one of "small" and "big"\n - sigmoid_loss: Set to true if nodes can belong to multiple classes\n - identity_dim: context embedding\n ' models.GeneralizedModel.__init__(self, **kwargs) if (aggregator_type == 'mean'): self.aggregator_cls = MeanAggregator elif (aggregator_type == 'seq'): self.aggregator_cls = SeqAggregator elif (aggregator_type == 'meanpool'): self.aggregator_cls = MeanPoolingAggregator elif (aggregator_type == 'maxpool'): self.aggregator_cls = MaxPoolingAggregator elif (aggregator_type == 'gcn'): self.aggregator_cls = GCNAggregator else: raise Exception('Unknown aggregator: ', self.aggregator_cls) self.inputs1 = placeholders['batch'] self.model_size = model_size self.adj_info = adj if (identity_dim > 0): self.embeds_context = tf.get_variable('node_embeddings', [features.shape[0], identity_dim]) else: self.embeds_context = None if (features is None): if (identity_dim == 0): raise Exception('Must have a positive value for identity feature dimension if no input features given.') self.features = self.embeds_context else: self.features = tf.Variable(tf.constant(features, dtype=tf.float32), trainable=False) if (not (self.embeds_context is None)): self.features = tf.concat([self.embeds_context, self.features], axis=1) self.degrees = degrees self.concat = concat self.num_classes = num_classes self.sigmoid_loss = sigmoid_loss self.dims = [((0 if (features is None) else features.shape[1]) + identity_dim)] self.dims.extend([layer_infos[0][i].output_dim for i in range(len(layer_infos[0]))]) self.batch_size = placeholders['batch_size'] self.placeholders = placeholders self.layer_infos = layer_infos self.num_relations = num_re dim_mult = (2 if self.concat else 1) self.relation_vectors = tf.Variable(glorot([num_re, (self.dims[(- 1)] * dim_mult)]), trainable=True, name='relation_vectors') self.attention_vec = tf.Variable(glorot([((self.dims[(- 1)] * dim_mult) * 2), 1])) self.optimizer = tf.train.AdamOptimizer(learning_rate=FLAGS.learning_rate) self.build() def build(self): (samples1_list, support_sizes1_list) = ([], []) for r_idx in range(self.num_relations): (samples1, support_sizes1) = self.sample(self.inputs1, self.layer_infos[r_idx]) samples1_list.append(samples1) support_sizes1_list.append(support_sizes1) num_samples = [layer_info.num_samples for layer_info in self.layer_infos[0]] self.outputs1_list = [] dim_mult = (2 if self.concat else 1) dim_mult = (dim_mult * 2) for r_idx in range(self.num_relations): (outputs1, self.aggregators) = self.aggregate(samples1_list[r_idx], [self.features], self.dims, num_samples, support_sizes1, concat=self.concat, model_size=self.model_size) self.relation_batch = tf.tile([tf.nn.embedding_lookup(self.relation_vectors, r_idx)], [self.batch_size, 1]) outputs1 = tf.concat([outputs1, self.relation_batch], 1) self.attention_weights = tf.matmul(outputs1, self.attention_vec) self.attention_weights = tf.tile(self.attention_weights, [1, (dim_mult * self.dims[(- 1)])]) outputs1 = tf.multiply(self.attention_weights, outputs1) self.outputs1_list += [outputs1] self.outputs1 = tf.stack(self.outputs1_list, 1) self.outputs1 = tf.reduce_sum(self.outputs1, axis=1, keepdims=False) self.outputs1 = tf.nn.l2_normalize(self.outputs1, 1) self.node_pred = layers.Dense((dim_mult * self.dims[(- 1)]), self.num_classes, dropout=self.placeholders['dropout'], act=(lambda x: x)) self.node_preds = self.node_pred(self.outputs1) self._loss() grads_and_vars = self.optimizer.compute_gradients(self.loss) clipped_grads_and_vars = [((tf.clip_by_value(grad, (- 5.0), 5.0) if (grad is not None) else None), var) for (grad, var) in grads_and_vars] (self.grad, _) = clipped_grads_and_vars[0] self.opt_op = self.optimizer.apply_gradients(clipped_grads_and_vars) self.preds = self.predict() def _loss(self): for aggregator in self.aggregators: for var in aggregator.vars.values(): self.loss += (FLAGS.weight_decay * tf.nn.l2_loss(var)) for var in self.node_pred.vars.values(): self.loss += (FLAGS.weight_decay * tf.nn.l2_loss(var)) if self.sigmoid_loss: self.loss += tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=self.node_preds, labels=self.placeholders['labels'])) else: self.loss += tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.node_preds, labels=self.placeholders['labels'])) tf.summary.scalar('loss', self.loss) def predict(self): if self.sigmoid_loss: return tf.nn.sigmoid(self.node_preds) else: return tf.nn.softmax(self.node_preds)
def calc_f1(y_true, y_pred): if (not FLAGS.sigmoid): y_true = np.argmax(y_true, axis=1) y_pred = np.argmax(y_pred, axis=1) else: y_pred[(y_pred > 0.5)] = 1 y_pred[(y_pred <= 0.5)] = 0 return (metrics.f1_score(y_true, y_pred, average='micro'), metrics.f1_score(y_true, y_pred, average='macro'))
def calc_auc(y_true, y_pred): return metrics.roc_auc_score(y_true, y_pred)
def evaluate(sess, model, minibatch_iter, size=None): t_test = time.time() (feed_dict_val, labels) = minibatch_iter.node_val_feed_dict(size) node_outs_val = sess.run([model.preds, model.loss], feed_dict=feed_dict_val) (mic, mac) = calc_f1(labels, node_outs_val[0]) auc = calc_auc(labels, node_outs_val[0]) return (node_outs_val[1], mic, mac, auc, (time.time() - t_test))
def incremental_evaluate(sess, model, minibatch_iter, size, test=False): t_test = time.time() finished = False val_losses = [] val_preds = [] labels = [] iter_num = 0 finished = False while (not finished): (feed_dict_val, batch_labels, finished, _) = minibatch_iter.incremental_node_val_feed_dict(size, iter_num, test=test) node_outs_val = sess.run([model.preds, model.loss], feed_dict=feed_dict_val) val_preds.append(node_outs_val[0]) labels.append(batch_labels) val_losses.append(node_outs_val[1]) iter_num += 1 val_preds = np.vstack(val_preds) labels = np.vstack(labels) f1_scores = calc_f1(labels, val_preds) auc_score = calc_auc(labels, val_preds) return (np.mean(val_losses), f1_scores[0], f1_scores[1], auc_score, (time.time() - t_test))
def construct_placeholders(num_classes): placeholders = {'labels': tf.placeholder(tf.float32, shape=(None, num_classes), name='labels'), 'batch': tf.placeholder(tf.int32, shape=None, name='batch1'), 'dropout': tf.placeholder_with_default(0.0, shape=(), name='dropout'), 'batch_size': tf.placeholder(tf.int32, name='batch_size')} return placeholders
def train(train_data, test_data=None): G = train_data[0] features = train_data[1] id_map = train_data[2] class_map = train_data[4] gs = train_data[5] num_relations = len(gs) if isinstance(list(class_map.values())[0], list): num_classes = len(list(class_map.values())[0]) else: num_classes = len(set(class_map.values())) if (not (features is None)): features = np.vstack([features, np.zeros((features.shape[1],))]) context_pairs = (train_data[3] if FLAGS.random_context else None) placeholders = construct_placeholders(num_classes) minibatch_list = [NodeMinibatchIterator(g, id_map, placeholders, class_map, num_classes, batch_size=FLAGS.batch_size, max_degree=FLAGS.max_degree, context_pairs=context_pairs) for g in gs] minibatch_main = NodeMinibatchIterator(G, id_map, placeholders, class_map, num_classes, batch_size=FLAGS.batch_size, max_degree=FLAGS.max_degree, context_pairs=context_pairs) adj_info_ph_list = [tf.placeholder(tf.int32, shape=minibatch_main.adj.shape) for i in range(num_relations)] adj_info_list = [tf.Variable(adj_info_ph, trainable=False, name='adj_info') for adj_info_ph in adj_info_ph_list] adj_info_main = adj_info_list[0] if (FLAGS.model == 'graphsage_mean'): sampler_list = [DistanceNeighborSampler(adj_info) for adj_info in adj_info_list] if (FLAGS.samples_3 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] for sampler in sampler_list] elif (FLAGS.samples_2 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] for sampler in sampler_list] else: hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] for sampler in sampler_list] model = SupervisedGraphconsis(num_classes, placeholders, features, adj_info_main, minibatch_main.deg, hete_layer_infos, model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.context_dim, num_re=num_relations, logging=True) elif (FLAGS.model == 'gcn'): sampler_list = [DistanceNeighborSampler(adj_info) for adj_info in adj_info_list] if (FLAGS.samples_3 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] for sampler in sampler_list] elif (FLAGS.samples_2 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] for sampler in sampler_list] else: hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] for sampler in sampler_list] model = SupervisedGraphconsis(num_classes, placeholders, features, adj_info_main, minibatch_main.deg, layer_infos=hete_layer_infos, aggregator_type='gcn', model_size=FLAGS.model_size, concat=False, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.context_dim, num_re=num_relations, logging=True) elif (FLAGS.model == 'graphsage_seq'): sampler_list = [DistanceNeighborSampler(adj_info) for adj_info in adj_info_list] if (FLAGS.samples_3 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] for sampler in sampler_list] elif (FLAGS.samples_2 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] for sampler in sampler_list] else: hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] for sampler in sampler_list] model = SupervisedGraphconsis(num_classes, placeholders, features, adj_info_main, minibatch_main.deg, layer_infos=hete_layer_infos, aggregator_type='seq', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.context_dim, num_re=num_relations, logging=True) elif (FLAGS.model == 'graphsage_maxpool'): sampler_list = [DistanceNeighborSampler(adj_info) for adj_info in adj_info_list] if (FLAGS.samples_3 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] for sampler in sampler_list] elif (FLAGS.samples_2 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] for sampler in sampler_list] else: hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] for sampler in sampler_list] model = SupervisedGraphconsis(num_classes, placeholders, features, adj_info_main, minibatch_main.deg, layer_infos=hete_layer_infos, aggregator_type='maxpool', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.context_dim, num_re=num_relations, logging=True) elif (FLAGS.model == 'graphsage_meanpool'): sampler_list = [DistanceNeighborSampler(adj_info) for adj_info in adj_info_list] if (FLAGS.samples_3 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] for sampler in sampler_list] elif (FLAGS.samples_2 != 0): hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] for sampler in sampler_list] else: hete_layer_infos = [[SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] for sampler in sampler_list] model = SupervisedGraphconsis(num_classes, placeholders, features, adj_info_main, minibatch_main.deg, layer_infos=hete_layer_infos, aggregator_type='meanpool', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.context_dim, num_re=num_relations, logging=True) else: raise Exception('Error: model name unrecognized.') config = tf.ConfigProto(log_device_placement=FLAGS.log_device_placement) config.gpu_options.allow_growth = True config.allow_soft_placement = True sess = tf.Session(config=config) merged = tf.summary.merge_all() sess.run(tf.global_variables_initializer(), feed_dict={adj_info_ph_list[i]: minibatch_list[i].adj for i in range(num_relations)}) total_steps = 0 avg_time = 0.0 epoch_val_costs = [] for epoch in range(FLAGS.epochs): minibatch_main.shuffle() iter = 0 print(('Epoch: %04d' % (epoch + 1))) epoch_val_costs.append(0) while (not minibatch_main.end()): (feed_dict, labels) = minibatch_main.next_minibatch_feed_dict() feed_dict.update({placeholders['dropout']: FLAGS.dropout}) t = time.time() outs = sess.run([merged, model.opt_op, model.loss, model.preds], feed_dict=feed_dict) train_cost = outs[2] if ((iter % FLAGS.validate_iter) == 0): if (FLAGS.validate_batch_size == (- 1)): ret = incremental_evaluate(sess, model, minibatch_main, FLAGS.batch_size) (val_cost, val_f1_mic, val_f1_mac, val_auc, duration) = ret else: ret = evaluate(sess, model, minibatch_main, FLAGS.validate_batch_size) (val_cost, val_f1_mic, val_f1_mac, val_auc, duration) = ret epoch_val_costs[(- 1)] += val_cost avg_time = ((((avg_time * total_steps) + time.time()) - t) / (total_steps + 1)) if ((total_steps % FLAGS.print_every) == 0): (train_f1_mic, train_f1_mac) = calc_f1(labels, outs[(- 1)]) print('Iter:', ('%04d' % iter), 'train_loss=', '{:.5f}'.format(train_cost), 'train_f1_mic=', '{:.5f}'.format(train_f1_mic), 'train_f1_mac=', '{:.5f}'.format(train_f1_mac), 'val_loss=', '{:.5f}'.format(val_cost), 'val_f1_mic=', '{:.5f}'.format(val_f1_mic), 'val_f1_mac=', '{:.5f}'.format(val_f1_mac), 'time=', '{:.5f}'.format(avg_time)) iter += 1 total_steps += 1 if (total_steps > FLAGS.max_total_steps): break if (total_steps > FLAGS.max_total_steps): break print('Optimization Finished!') ret = incremental_evaluate(sess, model, minibatch_main, FLAGS.batch_size) (val_cost, val_f1_mic, val_f1_mac, val_auc, duration) = ret print('Full validation stats:', 'loss=', '{:.5f}'.format(val_cost), 'f1_micro=', '{:.5f}'.format(val_f1_mic), 'f1_macro=', '{:.5f}'.format(val_f1_mac), 'auc=', '{:.5f}'.format(val_auc), 'time=', '{:.5f}'.format(duration))
def main(argv=None): print('Loading training data..') file_name = FLAGS.file_name train_perc = FLAGS.train_perc relations = ['net_rur', 'net_rtr', 'net_rsr'] train_data = load_data(FLAGS.train_prefix, file_name, relations, train_perc) print('Done loading training data..') train(train_data)
class MeanAggregator(Layer): '\n Aggregates via mean followed by matmul and non-linearity.\n ' def __init__(self, input_dim, output_dim, neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MeanAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([neigh_input_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_vecs = tf.nn.dropout(neigh_vecs, (1 - self.dropout)) self_vecs = tf.nn.dropout(self_vecs, (1 - self.dropout)) neigh_means = tf.reduce_mean(neigh_vecs, axis=1) from_neighs = tf.matmul(neigh_means, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class GCNAggregator(Layer): '\n Aggregates via mean followed by matmul and non-linearity.\n Same matmul parameters are used self vector and neighbor vectors.\n ' def __init__(self, input_dim, output_dim, neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(GCNAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['weights'] = glorot([neigh_input_dim, output_dim], name='neigh_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_vecs = tf.nn.dropout(neigh_vecs, (1 - self.dropout)) self_vecs = tf.nn.dropout(self_vecs, (1 - self.dropout)) means = tf.reduce_mean(tf.concat([neigh_vecs, tf.expand_dims(self_vecs, axis=1)], axis=1), axis=1) output = tf.matmul(means, self.vars['weights']) if self.bias: output += self.vars['bias'] return self.act(output)
class MaxPoolingAggregator(Layer): ' Aggregates via max-pooling over MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MaxPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 512 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 1024 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim)) neigh_h = tf.reduce_max(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class MeanPoolingAggregator(Layer): ' Aggregates via mean-pooling over MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(MeanPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 512 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 1024 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim)) neigh_h = tf.reduce_mean(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class TwoMaxLayerPoolingAggregator(Layer): ' Aggregates via pooling over two MLP functions.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(TwoMaxLayerPoolingAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim_1 = self.hidden_dim_1 = 512 hidden_dim_2 = self.hidden_dim_2 = 256 elif (model_size == 'big'): hidden_dim_1 = self.hidden_dim_1 = 1024 hidden_dim_2 = self.hidden_dim_2 = 512 self.mlp_layers = [] self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim_1, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) self.mlp_layers.append(Dense(input_dim=hidden_dim_1, output_dim=hidden_dim_2, act=tf.nn.relu, dropout=dropout, sparse_inputs=False, logging=self.logging)) with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim_2, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim def _call(self, inputs): (self_vecs, neigh_vecs) = inputs neigh_h = neigh_vecs dims = tf.shape(neigh_h) batch_size = dims[0] num_neighbors = dims[1] h_reshaped = tf.reshape(neigh_h, ((batch_size * num_neighbors), self.neigh_input_dim)) for l in self.mlp_layers: h_reshaped = l(h_reshaped) neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim_2)) neigh_h = tf.reduce_max(neigh_h, axis=1) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
class SeqAggregator(Layer): ' Aggregates via a standard LSTM.\n ' def __init__(self, input_dim, output_dim, model_size='small', neigh_input_dim=None, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(SeqAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat if (neigh_input_dim is None): neigh_input_dim = input_dim if (name is not None): name = ('/' + name) else: name = '' if (model_size == 'small'): hidden_dim = self.hidden_dim = 128 elif (model_size == 'big'): hidden_dim = self.hidden_dim = 256 with tf.variable_scope(((self.name + name) + '_vars')): self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights') self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim self.neigh_input_dim = neigh_input_dim self.cell = tf.contrib.rnn.BasicLSTMCell(self.hidden_dim) def _call(self, inputs): (self_vecs, neigh_vecs) = inputs dims = tf.shape(neigh_vecs) batch_size = dims[0] initial_state = self.cell.zero_state(batch_size, tf.float32) used = tf.sign(tf.reduce_max(tf.abs(neigh_vecs), axis=2)) length = tf.reduce_sum(used, axis=1) length = tf.maximum(length, tf.constant(1.0)) length = tf.cast(length, tf.int32) with tf.variable_scope(self.name) as scope: try: (rnn_outputs, rnn_states) = tf.nn.dynamic_rnn(self.cell, neigh_vecs, initial_state=initial_state, dtype=tf.float32, time_major=False, sequence_length=length) except ValueError: scope.reuse_variables() (rnn_outputs, rnn_states) = tf.nn.dynamic_rnn(self.cell, neigh_vecs, initial_state=initial_state, dtype=tf.float32, time_major=False, sequence_length=length) batch_size = tf.shape(rnn_outputs)[0] max_len = tf.shape(rnn_outputs)[1] out_size = int(rnn_outputs.get_shape()[2]) index = ((tf.range(0, batch_size) * max_len) + (length - 1)) flat = tf.reshape(rnn_outputs, [(- 1), out_size]) neigh_h = tf.gather(flat, index) from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights']) from_self = tf.matmul(self_vecs, self.vars['self_weights']) output = tf.add_n([from_self, from_neighs]) if (not self.concat): output = tf.add_n([from_self, from_neighs]) else: output = tf.concat([from_self, from_neighs], axis=1) if self.bias: output += self.vars['bias'] return self.act(output)
def uniform(shape, scale=0.05, name=None): 'Uniform init.' initial = tf.random_uniform(shape, minval=(- scale), maxval=scale, dtype=tf.float32) return tf.Variable(initial, name=name)
def glorot(shape, name=None): 'Glorot & Bengio (AISTATS 2010) init.' init_range = np.sqrt((6.0 / (shape[0] + shape[1]))) initial = tf.random_uniform(shape, minval=(- init_range), maxval=init_range, dtype=tf.float32) return tf.Variable(initial, name=name)
def zeros(shape, name=None): 'All zeros.' initial = tf.zeros(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def ones(shape, name=None): 'All ones.' initial = tf.ones(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def get_layer_uid(layer_name=''): 'Helper function, assigns unique layer IDs.' if (layer_name not in _LAYER_UIDS): _LAYER_UIDS[layer_name] = 1 return 1 else: _LAYER_UIDS[layer_name] += 1 return _LAYER_UIDS[layer_name]
class Layer(object): 'Base layer class. Defines basic API for all layer objects.\n Implementation inspired by keras (http://keras.io).\n # Properties\n name: String, defines the variable scope of the layer.\n logging: Boolean, switches Tensorflow histogram logging on/off\n\n # Methods\n _call(inputs): Defines computation graph of layer\n (i.e. takes input, returns output)\n __call__(inputs): Wrapper for _call()\n _log_vars(): Log all variables\n ' def __init__(self, **kwargs): allowed_kwargs = {'name', 'logging', 'model_size'} for kwarg in kwargs.keys(): assert (kwarg in allowed_kwargs), ('Invalid keyword argument: ' + kwarg) name = kwargs.get('name') if (not name): layer = self.__class__.__name__.lower() name = ((layer + '_') + str(get_layer_uid(layer))) self.name = name self.vars = {} logging = kwargs.get('logging', False) self.logging = logging self.sparse_inputs = False def _call(self, inputs): return inputs def __call__(self, inputs): with tf.name_scope(self.name): if (self.logging and (not self.sparse_inputs)): tf.summary.histogram((self.name + '/inputs'), inputs) outputs = self._call(inputs) if self.logging: tf.summary.histogram((self.name + '/outputs'), outputs) return outputs def _log_vars(self): for var in self.vars: tf.summary.histogram(((self.name + '/vars/') + var), self.vars[var])
class Dense(Layer): 'Dense layer.' def __init__(self, input_dim, output_dim, dropout=0.0, act=tf.nn.relu, placeholders=None, bias=True, featureless=False, sparse_inputs=False, **kwargs): super(Dense, self).__init__(**kwargs) self.dropout = dropout self.act = act self.featureless = featureless self.bias = bias self.input_dim = input_dim self.output_dim = output_dim self.sparse_inputs = sparse_inputs if sparse_inputs: self.num_features_nonzero = placeholders['num_features_nonzero'] with tf.variable_scope((self.name + '_vars')): self.vars['weights'] = tf.get_variable('weights', shape=(input_dim, output_dim), dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer(), regularizer=tf.contrib.layers.l2_regularizer(FLAGS.weight_decay)) if self.bias: self.vars['bias'] = zeros([output_dim], name='bias') if self.logging: self._log_vars() def _call(self, inputs): x = inputs x = tf.nn.dropout(x, (1 - self.dropout)) output = tf.matmul(x, self.vars['weights']) if self.bias: output += self.vars['bias'] return self.act(output)
def masked_logit_cross_entropy(preds, labels, mask): 'Logit cross-entropy loss with masking.' loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=preds, labels=labels) loss = tf.reduce_sum(loss, axis=1) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.maximum(tf.reduce_sum(mask), tf.constant([1.0])) loss *= mask return tf.reduce_mean(loss)
def masked_softmax_cross_entropy(preds, labels, mask): 'Softmax cross-entropy loss with masking.' loss = tf.nn.softmax_cross_entropy_with_logits(logits=preds, labels=labels) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.maximum(tf.reduce_sum(mask), tf.constant([1.0])) loss *= mask return tf.reduce_mean(loss)
def masked_l2(preds, actuals, mask): 'L2 loss with masking.' loss = tf.nn.l2(preds, actuals) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.reduce_mean(mask) loss *= mask return tf.reduce_mean(loss)
def masked_accuracy(preds, labels, mask): 'Accuracy with masking.' correct_prediction = tf.equal(tf.argmax(preds, 1), tf.argmax(labels, 1)) accuracy_all = tf.cast(correct_prediction, tf.float32) mask = tf.cast(mask, dtype=tf.float32) mask /= tf.reduce_mean(mask) accuracy_all *= mask return tf.reduce_mean(accuracy_all)
class UniformNeighborSampler(Layer): '\n Uniformly samples neighbors.\n Assumes that adj lists are padded with random re-sampling\n ' def __init__(self, adj_info, **kwargs): super(UniformNeighborSampler, self).__init__(**kwargs) self.adj_info = adj_info def _call(self, inputs): (ids, num_samples) = inputs adj_lists = tf.nn.embedding_lookup(self.adj_info, ids) adj_lists = tf.transpose(tf.random_shuffle(tf.transpose(adj_lists))) adj_lists = tf.slice(adj_lists, [0, 0], [(- 1), num_samples]) return adj_lists
class BipartiteEdgePredLayer(Layer): def __init__(self, input_dim1, input_dim2, placeholders, dropout=False, act=tf.nn.sigmoid, loss_fn='xent', neg_sample_weights=1.0, bias=False, bilinear_weights=False, **kwargs): '\n Basic class that applies skip-gram-like loss\n (i.e., dot product of node+target and node and negative samples)\n Args:\n bilinear_weights: use a bilinear weight for affinity calculation: u^T A v. If set to\n false, it is assumed that input dimensions are the same and the affinity will be \n based on dot product.\n ' super(BipartiteEdgePredLayer, self).__init__(**kwargs) self.input_dim1 = input_dim1 self.input_dim2 = input_dim2 self.act = act self.bias = bias self.eps = 1e-07 self.margin = 0.1 self.neg_sample_weights = neg_sample_weights self.bilinear_weights = bilinear_weights if dropout: self.dropout = placeholders['dropout'] else: self.dropout = 0.0 self.output_dim = 1 with tf.variable_scope((self.name + '_vars')): if bilinear_weights: self.vars['weights'] = tf.get_variable('pred_weights', shape=(input_dim1, input_dim2), dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer()) if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if (loss_fn == 'xent'): self.loss_fn = self._xent_loss elif (loss_fn == 'skipgram'): self.loss_fn = self._skipgram_loss elif (loss_fn == 'hinge'): self.loss_fn = self._hinge_loss if self.logging: self._log_vars() def affinity(self, inputs1, inputs2): ' Affinity score between batch of inputs1 and inputs2.\n Args:\n inputs1: tensor of shape [batch_size x feature_size].\n ' if self.bilinear_weights: prod = tf.matmul(inputs2, tf.transpose(self.vars['weights'])) self.prod = prod result = tf.reduce_sum((inputs1 * prod), axis=1) else: result = tf.reduce_sum((inputs1 * inputs2), axis=1) return result def neg_cost(self, inputs1, neg_samples, hard_neg_samples=None): ' For each input in batch, compute the sum of its affinity to negative samples.\n\n Returns:\n Tensor of shape [batch_size x num_neg_samples]. For each node, a list of affinities to\n negative samples is computed.\n ' if self.bilinear_weights: inputs1 = tf.matmul(inputs1, self.vars['weights']) neg_aff = tf.matmul(inputs1, tf.transpose(neg_samples)) return neg_aff def loss(self, inputs1, inputs2, neg_samples): ' negative sampling loss.\n Args:\n neg_samples: tensor of shape [num_neg_samples x input_dim2]. Negative samples for all\n inputs in batch inputs1.\n ' return self.loss_fn(inputs1, inputs2, neg_samples) def _xent_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) true_xent = tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(aff), logits=aff) negative_xent = tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.zeros_like(neg_aff), logits=neg_aff) loss = (tf.reduce_sum(true_xent) + (self.neg_sample_weights * tf.reduce_sum(negative_xent))) return loss def _skipgram_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) neg_cost = tf.log(tf.reduce_sum(tf.exp(neg_aff), axis=1)) loss = tf.reduce_sum((aff - neg_cost)) return loss def _hinge_loss(self, inputs1, inputs2, neg_samples, hard_neg_samples=None): aff = self.affinity(inputs1, inputs2) neg_aff = self.neg_cost(inputs1, neg_samples, hard_neg_samples) diff = tf.nn.relu(tf.subtract(neg_aff, (tf.expand_dims(aff, 1) - self.margin)), name='diff') loss = tf.reduce_sum(diff) self.neg_shape = tf.shape(neg_aff) return loss def weights_norm(self): return tf.nn.l2_norm(self.vars['weights'])
class SupervisedGraphsage(models.SampleAndAggregate): 'Implementation of supervised GraphSAGE.' def __init__(self, num_classes, placeholders, features, adj, degrees, layer_infos, concat=True, aggregator_type='mean', model_size='small', sigmoid_loss=False, identity_dim=0, **kwargs): '\n Args:\n - placeholders: Stanford TensorFlow placeholder object.\n - features: Numpy array with node features.\n - adj: Numpy array with adjacency lists (padded with random re-samples)\n - degrees: Numpy array with node degrees. \n - layer_infos: List of SAGEInfo namedtuples that describe the parameters of all \n the recursive layers. See SAGEInfo definition above.\n - concat: whether to concatenate during recursive iterations\n - aggregator_type: how to aggregate neighbor information\n - model_size: one of "small" and "big"\n - sigmoid_loss: Set to true if nodes can belong to multiple classes\n ' models.GeneralizedModel.__init__(self, **kwargs) if (aggregator_type == 'mean'): self.aggregator_cls = MeanAggregator elif (aggregator_type == 'seq'): self.aggregator_cls = SeqAggregator elif (aggregator_type == 'meanpool'): self.aggregator_cls = MeanPoolingAggregator elif (aggregator_type == 'maxpool'): self.aggregator_cls = MaxPoolingAggregator elif (aggregator_type == 'gcn'): self.aggregator_cls = GCNAggregator else: raise Exception('Unknown aggregator: ', self.aggregator_cls) self.inputs1 = placeholders['batch'] self.model_size = model_size self.adj_info = adj if (identity_dim > 0): self.embeds = tf.get_variable('node_embeddings', [adj.get_shape().as_list()[0], identity_dim]) else: self.embeds = None if (features is None): if (identity_dim == 0): raise Exception('Must have a positive value for identity feature dimension if no input features given.') self.features = self.embeds else: self.features = tf.Variable(tf.constant(features, dtype=tf.float32), trainable=False) if (not (self.embeds is None)): self.features = tf.concat([self.embeds, self.features], axis=1) self.degrees = degrees self.concat = concat self.num_classes = num_classes self.sigmoid_loss = sigmoid_loss self.dims = [((0 if (features is None) else features.shape[1]) + identity_dim)] self.dims.extend([layer_infos[i].output_dim for i in range(len(layer_infos))]) self.batch_size = placeholders['batch_size'] self.placeholders = placeholders self.layer_infos = layer_infos self.optimizer = tf.train.AdamOptimizer(learning_rate=FLAGS.learning_rate) self.build() def build(self): (samples1, support_sizes1) = self.sample(self.inputs1, self.layer_infos) num_samples = [layer_info.num_samples for layer_info in self.layer_infos] (self.outputs1, self.aggregators) = self.aggregate(samples1, [self.features], self.dims, num_samples, support_sizes1, concat=self.concat, model_size=self.model_size) dim_mult = (2 if self.concat else 1) self.outputs1 = tf.nn.l2_normalize(self.outputs1, 1) dim_mult = (2 if self.concat else 1) self.node_pred = layers.Dense((dim_mult * self.dims[(- 1)]), self.num_classes, dropout=self.placeholders['dropout'], act=(lambda x: x)) self.node_preds = self.node_pred(self.outputs1) self._loss() grads_and_vars = self.optimizer.compute_gradients(self.loss) clipped_grads_and_vars = [((tf.clip_by_value(grad, (- 5.0), 5.0) if (grad is not None) else None), var) for (grad, var) in grads_and_vars] (self.grad, _) = clipped_grads_and_vars[0] self.opt_op = self.optimizer.apply_gradients(clipped_grads_and_vars) self.preds = self.predict() def _loss(self): for aggregator in self.aggregators: for var in aggregator.vars.values(): self.loss += (FLAGS.weight_decay * tf.nn.l2_loss(var)) for var in self.node_pred.vars.values(): self.loss += (FLAGS.weight_decay * tf.nn.l2_loss(var)) if self.sigmoid_loss: self.loss += tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=self.node_preds, labels=self.placeholders['labels'])) else: self.loss += tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.node_preds, labels=self.placeholders['labels'])) tf.summary.scalar('loss', self.loss) def predict(self): if self.sigmoid_loss: return tf.nn.sigmoid(self.node_preds) else: return tf.nn.softmax(self.node_preds)
def calc_f1(y_true, y_pred): if (not FLAGS.sigmoid): y_true = np.argmax(y_true, axis=1) y_pred = np.argmax(y_pred, axis=1) else: y_pred[(y_pred > 0.5)] = 1 y_pred[(y_pred <= 0.5)] = 0 return (metrics.f1_score(y_true, y_pred, average='micro'), metrics.f1_score(y_true, y_pred, average='macro'))
def evaluate(sess, model, minibatch_iter, size=None): t_test = time.time() (feed_dict_val, labels) = minibatch_iter.node_val_feed_dict(size) node_outs_val = sess.run([model.preds, model.loss], feed_dict=feed_dict_val) (mic, mac) = calc_f1(labels, node_outs_val[0]) return (node_outs_val[1], mic, mac, (time.time() - t_test))
def log_dir(): log_dir = ((FLAGS.base_log_dir + '/sup-') + FLAGS.train_prefix.split('/')[(- 2)]) log_dir += '/{model:s}_{model_size:s}_{lr:0.4f}/'.format(model=FLAGS.model, model_size=FLAGS.model_size, lr=FLAGS.learning_rate) if (not os.path.exists(log_dir)): os.makedirs(log_dir) return log_dir
def incremental_evaluate(sess, model, minibatch_iter, size, test=False): t_test = time.time() finished = False val_losses = [] val_preds = [] labels = [] iter_num = 0 finished = False while (not finished): (feed_dict_val, batch_labels, finished, _) = minibatch_iter.incremental_node_val_feed_dict(size, iter_num, test=test) node_outs_val = sess.run([model.preds, model.loss], feed_dict=feed_dict_val) val_preds.append(node_outs_val[0]) labels.append(batch_labels) val_losses.append(node_outs_val[1]) iter_num += 1 val_preds = np.vstack(val_preds) labels = np.vstack(labels) f1_scores = calc_f1(labels, val_preds) return (np.mean(val_losses), f1_scores[0], f1_scores[1], (time.time() - t_test))
def construct_placeholders(num_classes): placeholders = {'labels': tf.placeholder(tf.float32, shape=(None, num_classes), name='labels'), 'batch': tf.placeholder(tf.int32, shape=None, name='batch1'), 'dropout': tf.placeholder_with_default(0.0, shape=(), name='dropout'), 'batch_size': tf.placeholder(tf.int32, name='batch_size')} return placeholders
def train(train_data, test_data=None): G = train_data[0] features = train_data[1] id_map = train_data[2] class_map = train_data[4] if isinstance(list(class_map.values())[0], list): num_classes = len(list(class_map.values())[0]) else: num_classes = len(set(class_map.values())) if (not (features is None)): features = np.vstack([features, np.zeros((features.shape[1],))]) context_pairs = (train_data[3] if FLAGS.random_context else None) placeholders = construct_placeholders(num_classes) minibatch = NodeMinibatchIterator(G, id_map, placeholders, class_map, num_classes, batch_size=FLAGS.batch_size, max_degree=FLAGS.max_degree, context_pairs=context_pairs) adj_info_ph = tf.placeholder(tf.int32, shape=minibatch.adj.shape) adj_info = tf.Variable(adj_info_ph, trainable=False, name='adj_info') if (FLAGS.model == 'graphsage_mean'): sampler = UniformNeighborSampler(adj_info) if (FLAGS.samples_3 != 0): layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2), SAGEInfo('node', sampler, FLAGS.samples_3, FLAGS.dim_2)] elif (FLAGS.samples_2 != 0): layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] else: layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1)] model = SupervisedGraphsage(num_classes, placeholders, features, adj_info, minibatch.deg, layer_infos, model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.identity_dim, logging=True) elif (FLAGS.model == 'gcn'): sampler = UniformNeighborSampler(adj_info) layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, (2 * FLAGS.dim_1)), SAGEInfo('node', sampler, FLAGS.samples_2, (2 * FLAGS.dim_2))] model = SupervisedGraphsage(num_classes, placeholders, features, adj_info, minibatch.deg, layer_infos=layer_infos, aggregator_type='gcn', model_size=FLAGS.model_size, concat=False, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.identity_dim, logging=True) elif (FLAGS.model == 'graphsage_seq'): sampler = UniformNeighborSampler(adj_info) layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] model = SupervisedGraphsage(num_classes, placeholders, features, adj_info, minibatch.deg, layer_infos=layer_infos, aggregator_type='seq', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.identity_dim, logging=True) elif (FLAGS.model == 'graphsage_maxpool'): sampler = UniformNeighborSampler(adj_info) layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] model = SupervisedGraphsage(num_classes, placeholders, features, adj_info, minibatch.deg, layer_infos=layer_infos, aggregator_type='maxpool', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.identity_dim, logging=True) elif (FLAGS.model == 'graphsage_meanpool'): sampler = UniformNeighborSampler(adj_info) layer_infos = [SAGEInfo('node', sampler, FLAGS.samples_1, FLAGS.dim_1), SAGEInfo('node', sampler, FLAGS.samples_2, FLAGS.dim_2)] model = SupervisedGraphsage(num_classes, placeholders, features, adj_info, minibatch.deg, layer_infos=layer_infos, aggregator_type='meanpool', model_size=FLAGS.model_size, sigmoid_loss=FLAGS.sigmoid, identity_dim=FLAGS.identity_dim, logging=True) else: raise Exception('Error: model name unrecognized.') config = tf.ConfigProto(log_device_placement=FLAGS.log_device_placement) config.gpu_options.allow_growth = True config.allow_soft_placement = True sess = tf.Session(config=config) merged = tf.summary.merge_all() summary_writer = tf.summary.FileWriter(log_dir(), sess.graph) sess.run(tf.global_variables_initializer(), feed_dict={adj_info_ph: minibatch.adj}) total_steps = 0 avg_time = 0.0 epoch_val_costs = [] train_adj_info = tf.assign(adj_info, minibatch.adj) val_adj_info = tf.assign(adj_info, minibatch.test_adj) for epoch in range(FLAGS.epochs): minibatch.shuffle() iter = 0 print(('Epoch: %04d' % (epoch + 1))) epoch_val_costs.append(0) while (not minibatch.end()): (feed_dict, labels) = minibatch.next_minibatch_feed_dict() feed_dict.update({placeholders['dropout']: FLAGS.dropout}) t = time.time() outs = sess.run([merged, model.opt_op, model.loss, model.preds], feed_dict=feed_dict) train_cost = outs[2] if ((iter % FLAGS.validate_iter) == 0): sess.run(val_adj_info.op) if (FLAGS.validate_batch_size == (- 1)): (val_cost, val_f1_mic, val_f1_mac, duration) = incremental_evaluate(sess, model, minibatch, FLAGS.batch_size) else: (val_cost, val_f1_mic, val_f1_mac, duration) = evaluate(sess, model, minibatch, FLAGS.validate_batch_size) sess.run(train_adj_info.op) epoch_val_costs[(- 1)] += val_cost if ((total_steps % FLAGS.print_every) == 0): summary_writer.add_summary(outs[0], total_steps) avg_time = ((((avg_time * total_steps) + time.time()) - t) / (total_steps + 1)) if ((total_steps % FLAGS.print_every) == 0): (train_f1_mic, train_f1_mac) = calc_f1(labels, outs[(- 1)]) print('Iter:', ('%04d' % iter), 'train_loss=', '{:.5f}'.format(train_cost), 'train_f1_mic=', '{:.5f}'.format(train_f1_mic), 'train_f1_mac=', '{:.5f}'.format(train_f1_mac), 'val_loss=', '{:.5f}'.format(val_cost), 'val_f1_mic=', '{:.5f}'.format(val_f1_mic), 'val_f1_mac=', '{:.5f}'.format(val_f1_mac), 'time=', '{:.5f}'.format(avg_time)) iter += 1 total_steps += 1 if (total_steps > FLAGS.max_total_steps): break if (total_steps > FLAGS.max_total_steps): break print('Optimization Finished!') sess.run(val_adj_info.op) (val_cost, val_f1_mic, val_f1_mac, duration) = incremental_evaluate(sess, model, minibatch, FLAGS.batch_size) print('Full validation stats:', 'loss=', '{:.5f}'.format(val_cost), 'f1_micro=', '{:.5f}'.format(val_f1_mic), 'f1_macro=', '{:.5f}'.format(val_f1_mac), 'time=', '{:.5f}'.format(duration)) with open((log_dir() + 'val_stats.txt'), 'w') as fp: fp.write('loss={:.5f} f1_micro={:.5f} f1_macro={:.5f} time={:.5f}'.format(val_cost, val_f1_mic, val_f1_mac, duration)) print("Writing test set stats to file (don't peak!)") (val_cost, val_f1_mic, val_f1_mac, duration) = incremental_evaluate(sess, model, minibatch, FLAGS.batch_size, test=True) with open((log_dir() + 'test_stats.txt'), 'w') as fp: fp.write('loss={:.5f} f1_micro={:.5f} f1_macro={:.5f}'.format(val_cost, val_f1_mic, val_f1_mac))
def main(argv=None): print('Loading training data..') train_data = load_data(FLAGS.train_prefix) print('Done loading training data..') train(train_data)
def calc_f1(y_true, y_pred): y_true = np.argmax(y_true, axis=1) y_pred = np.argmax(y_pred, axis=1) return (metrics.f1_score(y_true, y_pred, average='micro'), metrics.f1_score(y_true, y_pred, average='macro'))
def cal_acc(y_true, y_pred): y_true = np.argmax(y_true, axis=1) y_pred = np.argmax(y_pred, axis=1) return metrics.accuracy_score(y_true, y_pred)
class Model(object): def __init__(self, data_config, pretrain_data, args): self.model_type = 'hacud' self.adj_type = args.adj_type self.early_stop = args.early_stop self.pretrain_data = pretrain_data os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu self.n_nodes = data_config['n_nodes'] self.n_metapath = data_config['n_metapath'] self.n_class = data_config['n_class'] self.n_fold = args.n_fold self.n_fc = args.n_fc self.fc = eval(args.fc) self.reg = args.reg self.norm_adj = data_config['norm_adj'] self.features = data_config['features'] self.f_dim = self.features.shape[1] self.lr = args.lr self.emb_dim = args.embed_size self.batch_size = args.batch_size self.verbose = args.verbose '\n Create Placeholder for Input Data & Dropout.\n ' self.nodes = tf.placeholder(tf.int32, shape=(None,)) '\n Create Model Parameters (i.e., Initialize Weights).\n ' self.weights = self._init_weights() '\n Compute Graph-based Representations of all nodes\n ' self.n_embeddings = self._create_embedding() '\n Establish the representations of nodes in a batch.\n ' self.batch_embeddings = tf.nn.embedding_lookup(self.n_embeddings, self.nodes) self.label = tf.placeholder(tf.float32, shape=(None, self.n_class)) self.pred_label = self.pred(self.batch_embeddings) self.loss = self.create_loss(self.pred_label, self.label) self.opt = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.loss) def _init_weights(self): all_weights = dict() initializer = tf.contrib.layers.xavier_initializer() print('using xavier initialization') self.fc = ([self.emb_dim] + self.fc) all_weights['W'] = tf.Variable(initializer([self.f_dim, self.emb_dim]), name='W') all_weights['b'] = tf.Variable(initializer([1, self.emb_dim]), name='b') tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['W']) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['b']) for n in range(self.n_fc): all_weights[('W_%d' % n)] = tf.Variable(initializer([self.fc[n], self.fc[(n + 1)]]), name=('W_%d' % n)) all_weights[('b_%d' % n)] = tf.Variable(initializer([1, self.fc[(n + 1)]]), name=('b_%d' % n)) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('W_%d' % n)]) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('b_%d' % n)]) for n in range(self.n_metapath): all_weights[('W_rho_%d' % n)] = tf.Variable(initializer([self.f_dim, self.emb_dim]), name=('W_rho_%d' % n)) all_weights[('b_rho_%d' % n)] = tf.Variable(initializer([1, self.emb_dim]), name=('b_rho_%d' % n)) all_weights[('W_f_%d' % n)] = tf.Variable(initializer([(2 * self.emb_dim), self.emb_dim]), name=('W_f_%d' % n)) all_weights[('b_f_%d' % n)] = tf.Variable(initializer([1, self.emb_dim]), name=('b_f_%d' % n)) all_weights[('z_%d' % n)] = tf.Variable(initializer([1, (self.emb_dim * self.n_metapath)]), name=('z_%d' % n)) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('W_rho_%d' % n)]) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('b_rho_%d' % n)]) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('W_f_%d' % n)]) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('b_f_%d' % n)]) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights[('z_%d' % n)]) all_weights['W_f1'] = tf.Variable(initializer([(2 * self.emb_dim), self.emb_dim]), name='W_f1') all_weights['b_f1'] = tf.Variable(initializer([1, self.emb_dim]), name='b_f1') tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['W_f1']) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['b_f1']) all_weights['W_f2'] = tf.Variable(initializer([self.emb_dim, self.emb_dim]), name='W_f2') all_weights['b_f2'] = tf.Variable(initializer([1, self.emb_dim]), name='b_f2') tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['W_f2']) tf.add_to_collection(tf.GraphKeys.WEIGHTS, all_weights['b_f2']) return all_weights def _split_A_hat(self, X): A_fold_hat = [] fold_len = (self.n_nodes // self.n_fold) for i_fold in range(self.n_fold): start = (i_fold * fold_len) if (i_fold == (self.n_fold - 1)): end = self.n_nodes else: end = ((i_fold + 1) * fold_len) A_fold_hat.append(self._convert_sp_mat_to_sp_tensor(X[start:end])) return A_fold_hat def _create_embedding(self): A_fold_hat = {} for n in range(self.n_metapath): A_fold_hat[('%d' % n)] = self._split_A_hat(self.norm_adj[n]) embeddings = self.features embeddings = embeddings.astype(np.float32) h = (tf.matmul(embeddings, self.weights['W']) + self.weights['b']) embed_u = {} h_u = {} f_u = {} v_u = {} alp_u = {} alp_hat = {} f_tilde = {} for n in range(self.n_metapath): ' Graph Convolution ' embed_u[('%d' % n)] = [] for f in range(self.n_fold): embed_u[('%d' % n)].append(tf.sparse_tensor_dense_matmul(A_fold_hat[('%d' % n)][f], embeddings)) embed_u[('%d' % n)] = tf.concat(embed_u[('%d' % n)], 0) ' Feature Fusion ' h_u[('%d' % n)] = (tf.matmul(embed_u[('%d' % n)], self.weights[('W_rho_%d' % n)]) + self.weights[('b_rho_%d' % n)]) f_u[('%d' % n)] = tf.nn.relu((tf.matmul(tf.concat([h, h_u[('%d' % n)]], 1), self.weights[('W_f_%d' % n)]) + self.weights[('b_f_%d' % n)])) ' Feature Attention ' v_u[('%d' % n)] = tf.nn.relu((tf.matmul(tf.concat([h, f_u[('%d' % n)]], 1), self.weights['W_f1']) + self.weights['b_f1'])) alp_u[('%d' % n)] = tf.nn.relu((tf.matmul(v_u[('%d' % n)], self.weights['W_f2']) + self.weights['b_f2'])) alp_hat[('%d' % n)] = tf.nn.softmax(alp_u[('%d' % n)], axis=1) f_tilde[('%d' % n)] = tf.multiply(alp_hat[('%d' % n)], f_u[('%d' % n)]) ' Path Attention ' f_c = [] for n in range(self.n_metapath): f_c.append(f_tilde[('%d' % n)]) f_c = tf.concat(f_c, 1) for n in range(self.n_metapath): if (n == 0): beta = tf.matmul(f_c, tf.transpose(self.weights[('z_%d' % n)])) f = f_tilde[('%d' % n)] f = tf.expand_dims(f, (- 1)) else: beta = tf.concat([beta, tf.matmul(f_c, tf.transpose(self.weights[('z_%d' % n)]))], axis=1) f = tf.concat([f, tf.expand_dims(f_tilde[('%d' % n)], (- 1))], axis=2) beta_u = tf.nn.softmax(beta, axis=1) beta_u = tf.transpose(tf.expand_dims(beta_u, 0), (1, 0, 2)) e_u = tf.multiply(beta_u, f) e_u = tf.reduce_sum(e_u, axis=2) return e_u def pred(self, x): for n in range(self.n_fc): if (n == (self.n_fc - 1)): x = (tf.matmul(x, self.weights[('W_%d' % n)]) + self.weights[('b_%d' % n)]) else: x = tf.nn.relu((tf.matmul(x, self.weights[('W_%d' % n)]) + self.weights[('b_%d' % n)])) return x def create_ce_loss(self, x, y): ce_loss = tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(logits=x, labels=y), 0) return ce_loss def create_reg_loss(self): reg_loss = tf.add_n([tf.nn.l2_loss(tf.cast(v, tf.float32)) for v in tf.trainable_variables()]) return reg_loss def create_loss(self, x, y): self.ce_loss = self.create_ce_loss(x, y) self.reg_loss = self.create_reg_loss() loss = (self.ce_loss + (self.reg * self.reg_loss)) return loss def _convert_sp_mat_to_sp_tensor(self, X): coo = X.tocoo().astype(np.float32) indices = np.mat([coo.row, coo.col]).transpose() return tf.SparseTensor(indices, coo.data, coo.shape) def train(self, sess, nodes, labels): (_, batch_loss, batch_ce_loss, reg_loss) = sess.run([self.opt, self.loss, self.ce_loss, self.reg_loss], feed_dict={self.nodes: nodes, self.label: labels}) return (batch_loss, batch_ce_loss, reg_loss) def eval(self, sess, nodes, labels): (loss, ce_loss, reg_loss, pred_label) = sess.run([self.loss, self.ce_loss, self.reg_loss, self.pred_label], feed_dict={self.nodes: nodes, self.label: labels}) return (loss, ce_loss, reg_loss, pred_label)
def parse_args(): parser = argparse.ArgumentParser(description='Run HACUD.') parser.add_argument('--weights_path', nargs='?', default='', help='Store model path.') parser.add_argument('--data_path', nargs='?', default='../Data/', help='Input data path.') parser.add_argument('--proj_path', nargs='?', default='', help='Project path.') parser.add_argument('--dataset', nargs='?', default='dblp', help='Choose a dataset from {dblp, yelp}') parser.add_argument('--pretrain', type=int, default=0, help='0: No pretrain, -1: Pretrain with the learned embeddings, 1:Pretrain with stored models.') parser.add_argument('--verbose', type=int, default=1, help='Interval of evaluation.') parser.add_argument('--epoch', type=int, default=500, help='Number of epoch.') parser.add_argument('--embed_size', type=int, default=64, help='Embedding size.') parser.add_argument('--batch_size', type=int, default=64, help='Batch size.') parser.add_argument('--n_fold', type=int, default=20, help='number of fold.') parser.add_argument('--n_fc', type=int, default=4, help='number of fully-connected layers.') parser.add_argument('--fc', nargs='?', default='[32,16,8,4]', help='Output sizes of every layer') parser.add_argument('--lr', type=float, default=0.001, help='Learning rate.') parser.add_argument('--reg', type=float, default=0.001, help='Regularization ratio.') parser.add_argument('--model_type', nargs='?', default='ngcf', help='Specify the name of model (ngcf).') parser.add_argument('--adj_type', nargs='?', default='norm', help='Specify the type of the adjacency (laplacian) matrix from {plain, norm, mean}.') parser.add_argument('--alg_type', nargs='?', default='ngcf', help='Specify the type of the graph convolutional layer from {ngcf, gcn, gcmc}.') parser.add_argument('--save_flag', type=int, default=0, help='0: Disable model saver, 1: Activate model saver') parser.add_argument('--test_flag', nargs='?', default='part', help='Specify the test type from {part, full}, indicating whether the reference is done in mini-batch') parser.add_argument('--gpu', nargs='?', default='0') parser.add_argument('--early_stop', type=int, default=10) parser.add_argument('--report', type=int, default=0, help='0: Disable performance report w.r.t. sparsity levels, 1: Show performance report w.r.t. sparsity levels') return parser.parse_args()
class Player2Vec(Algorithm): def __init__(self, session, meta, nodes, class_size, gcn_output1, embedding, encoding): self.meta = meta self.nodes = nodes self.class_size = class_size self.gcn_output1 = gcn_output1 self.embedding = embedding self.encoding = encoding self.placeholders = {'a': tf.placeholder(tf.float32, [self.meta, self.nodes, self.nodes], 'adj'), 'x': tf.placeholder(tf.float32, [self.nodes, self.embedding], 'nxf'), 'batch_index': tf.placeholder(tf.int32, [None], 'index'), 't': tf.placeholder(tf.float32, [None, self.class_size], 'labels'), 'lr': tf.placeholder(tf.float32, [], 'learning_rate'), 'mom': tf.placeholder(tf.float32, [], 'momentum'), 'num_features_nonzero': tf.placeholder(tf.int32)} (loss, probabilities) = self.forward_propagation() (self.loss, self.probabilities) = (loss, probabilities) self.l2 = tf.contrib.layers.apply_regularization(tf.contrib.layers.l2_regularizer(0.01), tf.trainable_variables()) self.pred = tf.one_hot(tf.argmax(self.probabilities, 1), class_size) print(self.pred.shape) self.correct_prediction = tf.equal(tf.argmax(self.probabilities, 1), tf.argmax(self.placeholders['t'], 1)) self.accuracy = tf.reduce_mean(tf.cast(self.correct_prediction, 'float')) print('Forward propagation finished.') self.sess = session self.optimizer = tf.train.AdamOptimizer(self.placeholders['lr']) gradients = self.optimizer.compute_gradients((self.loss + self.l2)) capped_gradients = [(tf.clip_by_value(grad, (- 5.0), 5.0), var) for (grad, var) in gradients if (grad is not None)] self.train_op = self.optimizer.apply_gradients(capped_gradients) self.init = tf.global_variables_initializer() print('Backward propagation finished.') def forward_propagation(self): with tf.variable_scope('gcn'): gcn_emb = [] for i in range(self.meta): gcn_out = tf.reshape(GCN(self.placeholders, self.gcn_output1, self.embedding, self.encoding, index=i).embedding(), [1, (self.nodes * self.encoding)]) gcn_emb.append(gcn_out) gcn_emb = tf.concat(gcn_emb, 0) assert (gcn_emb.shape == [self.meta, (self.nodes * self.encoding)]) print('GCN embedding over!') with tf.variable_scope('attention'): gat_out = AttentionLayer.attention(inputs=gcn_emb, attention_size=1, v_type='tanh') gat_out = tf.reshape(gat_out, [self.nodes, self.encoding]) print('Embedding with attention over!') with tf.variable_scope('classification'): batch_data = tf.matmul(tf.one_hot(self.placeholders['batch_index'], self.nodes), gat_out) W = tf.get_variable(name='weights', shape=[self.encoding, self.class_size], initializer=tf.contrib.layers.xavier_initializer()) b = tf.get_variable(name='bias', shape=[1, self.class_size], initializer=tf.zeros_initializer()) tf.transpose(batch_data, perm=[0, 1]) logits = (tf.matmul(batch_data, W) + b) loss = tf.losses.sigmoid_cross_entropy(multi_class_labels=self.placeholders['t'], logits=logits) return (loss, tf.nn.sigmoid(logits)) def train(self, x, a, t, b, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict(x, a, t, b, learning_rate, momentum, self.placeholders) outs = self.sess.run([self.train_op, self.loss, self.accuracy, self.pred, self.probabilities], feed_dict=feed_dict) loss = outs[1] acc = outs[2] pred = outs[3] prob = outs[4] return (loss, acc, pred, prob) def test(self, x, a, t, b, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict(x, a, t, b, learning_rate, momentum, self.placeholders) (acc, pred, probabilities, tags) = self.sess.run([self.accuracy, self.pred, self.probabilities, self.correct_prediction], feed_dict=feed_dict) return (acc, pred, probabilities, tags)
def arg_parser(): parser = argparse.ArgumentParser() parser.add_argument('--seed', type=int, default=123, help='Random seed.') parser.add_argument('--dataset_str', type=str, default='dblp', help="['dblp','example']") parser.add_argument('--epoch_num', type=int, default=30, help='Number of epochs to train.') parser.add_argument('--batch_size', type=int, default=1000) parser.add_argument('--momentum', type=int, default=0.9) parser.add_argument('--learning_rate', default=0.001, help='the ratio of training set in whole dataset.') parser.add_argument('--hidden1', default=16, help='Number of units in GCN hidden layer 1.') parser.add_argument('--hidden2', default=16, help='Number of units in GCN hidden layer 2.') parser.add_argument('--gcn_output', default=4, help='gcn output size.') args = parser.parse_args() return args
def set_env(args): tf.reset_default_graph() np.random.seed(args.seed) tf.set_random_seed(args.seed)
def get_data(ix, int_batch, train_size): if ((ix + int_batch) >= train_size): ix = (train_size - int_batch) end = train_size else: end = (ix + int_batch) return (train_data[ix:end], train_label[ix:end])
def load_data(args): if (args.dataset_str == 'dblp'): (adj_list, features, train_data, train_label, test_data, test_label) = load_data_dblp() node_size = features.shape[0] node_embedding = features.shape[1] class_size = train_label.shape[1] train_size = len(train_data) paras = [node_size, node_embedding, class_size, train_size] return (adj_list, features, train_data, train_label, test_data, test_label, paras)
def train(args, adj_list, features, train_data, train_label, test_data, test_label, paras): with tf.Session() as sess: adj_data = [normalize_adj(adj) for adj in adj_list] meta_size = len(adj_list) net = Player2Vec(session=sess, class_size=paras[2], gcn_output1=args.hidden1, meta=meta_size, nodes=paras[0], embedding=paras[1], encoding=args.gcn_output) sess.run(tf.global_variables_initializer()) t_start = time.clock() for epoch in range(args.epoch_num): train_loss = 0 train_acc = 0 count = 0 for index in range(0, paras[3], args.batch_size): (batch_data, batch_label) = get_data(index, args.batch_size, paras[3]) (loss, acc, pred, prob) = net.train(features, adj_data, batch_label, batch_data, args.learning_rate, args.momentum) print('batch loss: {:.4f}, batch acc: {:.4f}'.format(loss, acc)) train_loss += loss train_acc += acc count += 1 train_loss = (train_loss / count) train_acc = (train_acc / count) print('epoch{:d} : train_loss: {:.4f}, train_acc: {:.4f}'.format(epoch, train_loss, train_acc)) t_end = time.clock() print('train time=', '{:.5f}'.format((t_end - t_start))) print('Train end!') (test_acc, test_pred, test_probabilities, test_tags) = net.test(features, adj_data, test_label, test_data) print('test acc:', test_acc)
class SemiGNN(Algorithm): def __init__(self, session, nodes, class_size, semi_encoding1, semi_encoding2, semi_encoding3, init_emb_size, meta, ul, alpha, lamtha): self.nodes = nodes self.meta = meta self.class_size = class_size self.semi_encoding1 = semi_encoding1 self.semi_encoding2 = semi_encoding2 self.semi_encoding3 = semi_encoding3 self.init_emb_size = init_emb_size self.ul = ul self.alpha = alpha self.lamtha = lamtha self.placeholders = {'a': tf.placeholder(tf.float32, [self.meta, self.nodes, self.nodes], 'adj'), 'u_i': tf.placeholder(tf.float32, [None], 'u_i'), 'u_j': tf.placeholder(tf.float32, [None], 'u_j'), 'batch_index': tf.placeholder(tf.int32, [None], 'index'), 'sup_label': tf.placeholder(tf.float32, [None, self.class_size], 'sup_label'), 'graph_label': tf.placeholder(tf.float32, [None, 1], 'graph_label'), 'lr': tf.placeholder(tf.float32, [], 'learning_rate'), 'mom': tf.placeholder(tf.float32, [], 'momentum'), 'num_features_nonzero': tf.placeholder(tf.int32)} (loss, probabilities, pred) = self.forward_propagation() (self.loss, self.probabilities, self.pred) = (loss, probabilities, pred) self.l2 = tf.contrib.layers.apply_regularization(tf.contrib.layers.l2_regularizer(0.01), tf.trainable_variables()) print(self.pred.shape) self.correct_prediction = tf.equal(tf.argmax(self.probabilities, 1), tf.argmax(self.placeholders['sup_label'], 1)) self.accuracy = tf.reduce_mean(tf.cast(self.correct_prediction, 'float')) print('Forward propagation finished.') self.sess = session self.optimizer = tf.train.AdamOptimizer(self.placeholders['lr']) gradients = self.optimizer.compute_gradients((self.loss + (self.lamtha * self.l2))) capped_gradients = [(tf.clip_by_value(grad, (- 5.0), 5.0), var) for (grad, var) in gradients if (grad is not None)] self.train_op = self.optimizer.apply_gradients(capped_gradients) self.init = tf.global_variables_initializer() print('Backward propagation finished.') def forward_propagation(self): with tf.variable_scope('node_level_attention', reuse=tf.AUTO_REUSE): h1 = [] for i in range(self.meta): emb = tf.get_variable(name='init_embedding', shape=[self.nodes, self.init_emb_size], initializer=tf.contrib.layers.xavier_initializer()) h = AttentionLayer.node_attention(inputs=emb, adj=self.placeholders['a'][i]) h = tf.reshape(h, [self.nodes, emb.shape[1]]) h1.append(h) h1 = tf.concat(h1, 0) h1 = tf.reshape(h1, [self.meta, self.nodes, self.init_emb_size]) print('Node_level attention over!') with tf.variable_scope('view_level_attention'): h2 = AttentionLayer.view_attention(inputs=h1, layer_size=2, meta=self.meta, encoding1=self.semi_encoding1, encoding2=self.semi_encoding2) h2 = tf.reshape(h2, [self.nodes, (self.semi_encoding2 * self.meta)]) print('View_level attention over!') with tf.variable_scope('MLP'): a_u = tf.layers.dense(inputs=h2, units=self.semi_encoding3, activation=None) with tf.variable_scope('loss'): labeled_a_u = tf.matmul(tf.one_hot(self.placeholders['batch_index'], self.nodes), a_u) theta = tf.get_variable(name='theta', shape=[self.semi_encoding3, self.class_size], initializer=tf.contrib.layers.xavier_initializer()) logits = tf.matmul(labeled_a_u, theta) prob = tf.nn.sigmoid(logits) pred = tf.one_hot(tf.argmax(prob, 1), self.class_size) loss1 = ((- (1 / self.ul)) * tf.reduce_sum((self.placeholders['sup_label'] * tf.log(tf.nn.softmax(logits))))) u_i_embedding = tf.nn.embedding_lookup(a_u, tf.cast(self.placeholders['u_i'], dtype=tf.int32)) u_j_embedding = tf.nn.embedding_lookup(a_u, tf.cast(self.placeholders['u_j'], dtype=tf.int32)) inner_product = tf.reduce_sum((u_i_embedding * u_j_embedding), axis=1) loss2 = (- tf.reduce_mean(tf.log_sigmoid((self.placeholders['graph_label'] * inner_product)))) loss = ((self.alpha * loss1) + ((1 - self.alpha) * loss2)) return (loss, prob, pred) def train(self, a, u_i, u_j, batch_graph_label, batch_data, batch_sup_label, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict_semi(a, u_i, u_j, batch_graph_label, batch_data, batch_sup_label, learning_rate, momentum, self.placeholders) outs = self.sess.run([self.train_op, self.loss, self.accuracy, self.pred, self.probabilities], feed_dict=feed_dict) loss = outs[1] acc = outs[2] pred = outs[3] prob = outs[4] return (loss, acc, pred, prob) def test(self, a, u_i, u_j, batch_graph_label, batch_data, batch_sup_label, learning_rate=0.01, momentum=0.9): feed_dict = utils.construct_feed_dict_semi(a, u_i, u_j, batch_graph_label, batch_data, batch_sup_label, learning_rate, momentum, self.placeholders) (acc, pred, probabilities, tags) = self.sess.run([self.accuracy, self.pred, self.probabilities, self.correct_prediction], feed_dict=feed_dict) return (acc, pred, probabilities, tags)
def arg_parser(): parser = argparse.ArgumentParser() parser.add_argument('--seed', type=int, default=123, help='Random seed.') parser.add_argument('--dataset_str', type=str, default='example', help="['dblp','example']") parser.add_argument('--epoch_num', type=int, default=30, help='Number of epochs to train.') parser.add_argument('--batch_size', type=int, default=1000) parser.add_argument('--momentum', type=int, default=0.9) parser.add_argument('--learning_rate', default=0.001, help='the ratio of training set in whole dataset.') parser.add_argument('--init_emb_size', default=4, help='initial node embedding size') parser.add_argument('--semi_encoding1', default=3, help='the first view attention layer unit number') parser.add_argument('--semi_encoding2', default=2, help='the second view attention layer unit number') parser.add_argument('--semi_encoding3', default=4, help='one-layer perceptron units') parser.add_argument('--Ul', default=8, help='labeled users number') parser.add_argument('--alpha', default=0.5, help='loss alpha') parser.add_argument('--lamtha', default=0.5, help='loss lamtha') args = parser.parse_args() return args
def set_env(args): tf.reset_default_graph() np.random.seed(args.seed) tf.set_random_seed(args.seed)
def get_data(ix, int_batch, train_size): if ((ix + int_batch) >= train_size): ix = (train_size - int_batch) end = train_size else: end = (ix + int_batch) return (train_data[ix:end], train_label[ix:end])
def load_data(args): if (args.dataset_str == 'example'): (adj_list, features, train_data, train_label, test_data, test_label) = load_example_semi() node_size = features.shape[0] node_embedding = features.shape[1] class_size = train_label.shape[1] train_size = len(train_data) paras = [node_size, node_embedding, class_size, train_size] return (adj_list, features, train_data, train_label, test_data, test_label, paras)
def train(args, adj_list, features, train_data, train_label, test_data, test_label, paras): with tf.Session() as sess: adj_nodelists = [matrix_to_adjlist(adj, pad=False) for adj in adj_list] meta_size = len(adj_list) pairs = [random_walks(adj_nodelists[i], 2, 3) for i in range(meta_size)] net = SemiGNN(session=sess, class_size=paras[2], semi_encoding1=args.semi_encoding1, semi_encoding2=args.semi_encoding2, semi_encoding3=args.semi_encoding3, meta=meta_size, nodes=paras[0], init_emb_size=args.init_emb_size, ul=args.batch_size, alpha=args.alpha, lamtha=args.lamtha) adj_data = [pairs_to_matrix(p, paras[0]) for p in pairs] u_i = [] u_j = [] for (adj_nodelist, p) in zip(adj_nodelists, pairs): (u_i_t, u_j_t, graph_label) = get_negative_sampling(p, adj_nodelist) u_i.append(u_i_t) u_j.append(u_j_t) u_i = np.concatenate(np.array(u_i)) u_j = np.concatenate(np.array(u_j)) sess.run(tf.global_variables_initializer()) t_start = time.clock() for epoch in range(args.epoch_num): train_loss = 0 train_acc = 0 count = 0 for index in range(0, paras[3], args.batch_size): (batch_data, batch_sup_label) = get_data(index, args.batch_size, paras[3]) (loss, acc, pred, prob) = net.train(adj_data, u_i, u_j, graph_label, batch_data, batch_sup_label, args.learning_rate, args.momentum) print('batch loss: {:.4f}, batch acc: {:.4f}'.format(loss, acc)) train_loss += loss train_acc += acc count += 1 train_loss = (train_loss / count) train_acc = (train_acc / count) print('epoch{:d} : train_loss: {:.4f}, train_acc: {:.4f}'.format(epoch, train_loss, train_acc)) t_end = time.clock() print('train time=', '{:.5f}'.format((t_end - t_start))) print('Train end!') (test_acc, test_pred, test_probabilities, test_tags) = net.test(adj_data, u_i, u_j, graph_label, test_data, test_label, args.learning_rate, args.momentum) print('test acc:', test_acc)
class Algorithm(object): def __init__(self, **kwargs): self.nodes = None def forward_propagation(self): pass def save(self, sess=None): if (not sess): raise AttributeError('TensorFlow session not provided.') saver = tf.train.Saver() save_path = saver.save(sess, ('tmp/%s.ckpt' % 'temp')) print(('Model saved in file: %s' % save_path)) def load(self, sess=None): if (not sess): raise AttributeError('TensorFlow session not provided.') saver = tf.train.Saver() save_path = ('tmp/%s.ckpt' % 'temp') saver.restore(sess, save_path) print(('Model restored from file: %s' % save_path))
def uniform(shape, scale=0.05, name=None): 'Uniform init.' initial = tf.random_uniform(shape, minval=(- scale), maxval=scale, dtype=tf.float32) return tf.Variable(initial, name=name)
def glorot(shape, name=None): 'Glorot & Bengio (AISTATS 2010) init.' init_range = np.sqrt((6.0 / (shape[0] + shape[1]))) initial = tf.random_uniform(shape, minval=(- init_range), maxval=init_range, dtype=tf.float32) return tf.Variable(initial, name=name)
def zeros(shape, name=None): 'All zeros.' initial = tf.zeros(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def ones(shape, name=None): 'All ones.' initial = tf.ones(shape, dtype=tf.float32) return tf.Variable(initial, name=name)
def get_layer_uid(layer_name=''): 'Helper function, assigns unique layer IDs.' if (layer_name not in _LAYER_UIDS): _LAYER_UIDS[layer_name] = 1 return 1 else: _LAYER_UIDS[layer_name] += 1 return _LAYER_UIDS[layer_name]
def sparse_dropout(x, keep_prob, noise_shape): 'Dropout for sparse tensors.' random_tensor = keep_prob random_tensor += tf.random_uniform(noise_shape) dropout_mask = tf.cast(tf.floor(random_tensor), dtype=tf.bool) pre_out = tf.sparse_retain(x, dropout_mask) return (pre_out * (1.0 / keep_prob))
def dot(x, y, sparse=False): 'Wrapper for tf.matmul (sparse vs dense).' if sparse: res = tf.sparse_tensor_dense_matmul(x, y) else: res = tf.matmul(x, y) return res
class Layer(object): 'Base layer class. Defines basic API for all layer objects.\n Implementation inspired by keras (http://keras.io).\n\n # Properties\n name: String, defines the variable scope of the layer.\n logging: Boolean, switches Tensorflow histogram logging on/off\n\n # Methods\n _call(inputs): Defines computation graph of layer\n (i.e. takes input, returns output)\n __call__(inputs): Wrapper for _call()\n _log_vars(): Log all variables\n ' def __init__(self, **kwargs): allowed_kwargs = {'name', 'logging'} for kwarg in kwargs.keys(): assert (kwarg in allowed_kwargs), ('Invalid keyword argument: ' + kwarg) name = kwargs.get('name') if (not name): layer = self.__class__.__name__.lower() name = ((layer + '_') + str(get_layer_uid(layer))) self.name = name self.vars = {} logging = kwargs.get('logging', False) self.logging = logging self.sparse_inputs = False def _call(self, inputs): return inputs def _call(self, inputs, adj_info): return inputs def __call__(self, inputs): with tf.name_scope(self.name): if (self.logging and (not self.sparse_inputs)): tf.summary.histogram((self.name + '/inputs'), inputs) outputs = self._call(inputs) if self.logging: tf.summary.histogram((self.name + '/outputs'), outputs) return outputs def _log_vars(self): for var in self.vars: tf.summary.histogram(((self.name + '/vars/') + var), self.vars[var])
class GraphConvolution(Layer): 'Graph convolution layer.' def __init__(self, input_dim, output_dim, placeholders, index=0, dropout=0.0, sparse_inputs=False, act=tf.nn.relu, bias=False, featureless=False, norm=False, **kwargs): super(GraphConvolution, self).__init__(**kwargs) self.dropout = dropout self.act = act self.support = placeholders['a'] self.sparse_inputs = sparse_inputs self.featureless = featureless self.bias = bias self.norm = norm self.index = index self.num_features_nonzero = placeholders['num_features_nonzero'] with tf.variable_scope((self.name + '_vars')): for i in range(1): self.vars[('weights_' + str(i))] = glorot([input_dim, output_dim], name=('weights_' + str(i))) if self.bias: self.vars['bias'] = zeros([output_dim], name='bias') if self.logging: self._log_vars() def _call(self, inputs): x = inputs if self.sparse_inputs: x = sparse_dropout(x, (1 - self.dropout), self.num_features_nonzero) else: x = tf.nn.dropout(x, (1 - self.dropout)) supports = list() for i in range(1): if (not self.featureless): pre_sup = dot(x, self.vars[('weights_' + str(i))], sparse=self.sparse_inputs) else: pre_sup = self.vars[('weights_' + str(i))] support = dot(self.support[self.index], pre_sup, sparse=False) supports.append(support) output = tf.add_n(supports) axis = list(range((len(output.get_shape()) - 1))) (mean, variance) = tf.nn.moments(output, axis) scale = None offset = None variance_epsilon = 0.001 output = tf.nn.batch_normalization(output, mean, variance, offset, scale, variance_epsilon) if self.bias: output += self.vars['bias'] if self.norm: return tf.nn.l2_normalize(self.act(output), axis=None, epsilon=1e-12) return self.act(output)
class AttentionLayer(Layer): ' AttentionLayer is a function f : hkey × Hval → hval which maps\n a feature vector hkey and the set of candidates’ feature vectors\n Hval to an weighted sum of elements in Hval.\n ' def attention(inputs, attention_size, v_type=None, return_weights=False, bias=True, joint_type='weighted_sum', multi_view=True): if multi_view: inputs = tf.expand_dims(inputs, 0) hidden_size = inputs.shape[(- 1)].value w_omega = tf.Variable(tf.random_normal([hidden_size, attention_size], stddev=0.1)) b_omega = tf.Variable(tf.random_normal([attention_size], stddev=0.1)) u_omega = tf.Variable(tf.random_normal([attention_size], stddev=0.1)) with tf.name_scope('v'): v = tf.tensordot(inputs, w_omega, axes=1) if (bias is True): v += b_omega if (v_type is 'tanh'): v = tf.tanh(v) if (v_type is 'relu'): v = tf.nn.relu(v) vu = tf.tensordot(v, u_omega, axes=1, name='vu') weights = tf.nn.softmax(vu, name='alphas') if (joint_type is 'weighted_sum'): output = tf.reduce_sum((inputs * tf.expand_dims(weights, (- 1))), 1) if (joint_type is 'concatenation'): output = tf.concat((inputs * tf.expand_dims(weights, (- 1))), 2) if (not return_weights): return output else: return (output, weights) def node_attention(inputs, adj, return_weights=False): hidden_size = inputs.shape[(- 1)].value H_v = tf.Variable(tf.random_normal([hidden_size, 1], stddev=0.1)) zero = tf.constant(0, dtype=tf.float32) where = tf.not_equal(adj, zero) indices = tf.where(where) values = tf.gather_nd(adj, indices) adj = tf.SparseTensor(indices=indices, values=values, dense_shape=adj.shape) with tf.name_scope('v'): v = (adj * tf.squeeze(tf.tensordot(inputs, H_v, axes=1))) weights = tf.sparse_softmax(v, name='alphas') output = tf.sparse_tensor_dense_matmul(weights, inputs) if (not return_weights): return output else: return (output, weights) def view_attention(inputs, encoding1, encoding2, layer_size, meta, return_weights=False): h = inputs encoding = [encoding1, encoding2] for l in range(layer_size): v = [] for i in range(meta): input = h[i] v_i = tf.layers.dense(inputs=input, units=encoding[l], activation=tf.nn.relu) v.append(v_i) h = v h = tf.concat(h, 0) h = tf.reshape(h, [meta, inputs[0].shape[0].value, encoding2]) phi = tf.Variable(tf.random_normal([encoding2], stddev=0.1)) weights = tf.nn.softmax((h * phi), name='alphas') output = tf.reshape((h * weights), [1, ((inputs[0].shape[0] * encoding2) * meta)]) if (not return_weights): return output else: return (output, weights) def scaled_dot_product_attention(q, k, v, mask): qk = tf.matmul(q, k, transpose_b=True) dk = tf.cast(tf.shape(k)[(- 1)], tf.float32) scaled_attention = (qk / tf.math.sqrt(dk)) if (mask is not None): scaled_attention += 1 weights = tf.nn.softmax(scaled_attention, axis=(- 1)) output = tf.matmul(weights, v) return (output, weights)
class ConcatenationAggregator(Layer): "This layer equals to the equation (3) in\n paper 'Spam Review Detection with Graph Convolutional Networks.'\n " def __init__(self, input_dim, output_dim, review_item_adj, review_user_adj, review_vecs, user_vecs, item_vecs, dropout=0.0, act=tf.nn.relu, name=None, concat=False, **kwargs): super(ConcatenationAggregator, self).__init__(**kwargs) self.review_item_adj = review_item_adj self.review_user_adj = review_user_adj self.review_vecs = review_vecs self.user_vecs = user_vecs self.item_vecs = item_vecs self.dropout = dropout self.act = act self.concat = concat if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['con_agg_weights'] = glorot([input_dim, output_dim], name='con_agg_weights') if self.logging: self._log_vars() self.input_dim = input_dim self.output_dim = output_dim def _call(self, inputs): review_vecs = tf.nn.dropout(self.review_vecs, (1 - self.dropout)) user_vecs = tf.nn.dropout(self.user_vecs, (1 - self.dropout)) item_vecs = tf.nn.dropout(self.item_vecs, (1 - self.dropout)) ri = tf.nn.embedding_lookup(item_vecs, tf.cast(self.review_item_adj, dtype=tf.int32)) ri = tf.transpose(tf.random_shuffle(tf.transpose(ri))) ru = tf.nn.embedding_lookup(user_vecs, tf.cast(self.review_user_adj, dtype=tf.int32)) ru = tf.transpose(tf.random_shuffle(tf.transpose(ru))) concate_vecs = tf.concat([review_vecs, ru, ri], axis=1) output = tf.matmul(concate_vecs, self.vars['con_agg_weights']) return self.act(output)
class AttentionAggregator(Layer): "This layer equals to equation (5) and equation (8) in\n paper 'Spam Review Detection with Graph Convolutional Networks.'\n " def __init__(self, input_dim1, input_dim2, output_dim, hid_dim, user_review_adj, user_item_adj, item_review_adj, item_user_adj, review_vecs, user_vecs, item_vecs, dropout=0.0, bias=False, act=tf.nn.relu, name=None, concat=False, **kwargs): super(AttentionAggregator, self).__init__(**kwargs) self.dropout = dropout self.bias = bias self.act = act self.concat = concat self.user_review_adj = user_review_adj self.user_item_adj = user_item_adj self.item_review_adj = item_review_adj self.item_user_adj = item_user_adj self.review_vecs = review_vecs self.user_vecs = user_vecs self.item_vecs = item_vecs if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['user_weights'] = glorot([input_dim1, hid_dim], name='user_weights') self.vars['item_weights'] = glorot([input_dim2, hid_dim], name='item_weights') self.vars['concate_user_weights'] = glorot([hid_dim, output_dim], name='user_weights') self.vars['concate_item_weights'] = glorot([hid_dim, output_dim], name='item_weights') if self.bias: self.vars['bias'] = zeros([self.output_dim], name='bias') if self.logging: self._log_vars() self.input_dim1 = input_dim1 self.input_dim2 = input_dim2 self.output_dim = output_dim def _call(self, inputs): review_vecs = tf.nn.dropout(self.review_vecs, (1 - self.dropout)) user_vecs = tf.nn.dropout(self.user_vecs, (1 - self.dropout)) item_vecs = tf.nn.dropout(self.item_vecs, (1 - self.dropout)) ur = tf.nn.embedding_lookup(review_vecs, tf.cast(self.user_review_adj, dtype=tf.int32)) ur = tf.transpose(tf.random_shuffle(tf.transpose(ur))) ri = tf.nn.embedding_lookup(item_vecs, tf.cast(self.user_item_adj, dtype=tf.int32)) ri = tf.transpose(tf.random_shuffle(tf.transpose(ri))) ir = tf.nn.embedding_lookup(review_vecs, tf.cast(self.item_review_adj, dtype=tf.int32)) ir = tf.transpose(tf.random_shuffle(tf.transpose(ir))) ru = tf.nn.embedding_lookup(user_vecs, tf.cast(self.item_user_adj, dtype=tf.int32)) ru = tf.transpose(tf.random_shuffle(tf.transpose(ru))) concate_user_vecs = tf.concat([ur, ri], axis=2) concate_item_vecs = tf.concat([ir, ru], axis=2) s1 = tf.shape(concate_user_vecs) s2 = tf.shape(concate_item_vecs) concate_user_vecs = tf.reshape(concate_user_vecs, [s1[0], (s1[1] * s1[2])]) concate_item_vecs = tf.reshape(concate_item_vecs, [s2[0], (s2[1] * s2[2])]) (concate_user_vecs, _) = AttentionLayer.scaled_dot_product_attention(q=user_vecs, k=user_vecs, v=concate_user_vecs, mask=None) (concate_item_vecs, _) = AttentionLayer.scaled_dot_product_attention(q=item_vecs, k=item_vecs, v=concate_item_vecs, mask=None) user_output = tf.matmul(concate_user_vecs, self.vars['user_weights']) item_output = tf.matmul(concate_item_vecs, self.vars['item_weights']) if self.bias: user_output += self.vars['bias'] item_output += self.vars['bias'] user_output = self.act(user_output) item_output = self.act(item_output) if self.concat: user_output = tf.matmul(user_output, self.vars['concate_user_weights']) item_output = tf.matmul(item_output, self.vars['concate_item_weights']) user_output = tf.concat([user_vecs, user_output], axis=1) item_output = tf.concat([item_vecs, item_output], axis=1) return (user_output, item_output)
class GASConcatenation(Layer): 'GCN-based Anti-Spam(GAS) layer for concatenation of comment embedding learned by GCN from the Comment Graph\n and other embeddings learned in previous operations.\n ' def __init__(self, review_item_adj, review_user_adj, review_vecs, item_vecs, user_vecs, homo_vecs, name=None, **kwargs): super(GASConcatenation, self).__init__(**kwargs) self.review_item_adj = review_item_adj self.review_user_adj = review_user_adj self.review_vecs = review_vecs self.user_vecs = user_vecs self.item_vecs = item_vecs self.homo_vecs = homo_vecs if (name is not None): name = ('/' + name) else: name = '' if self.logging: self._log_vars() def _call(self, inputs): ri = tf.nn.embedding_lookup(self.item_vecs, tf.cast(self.review_item_adj, dtype=tf.int32)) ru = tf.nn.embedding_lookup(self.user_vecs, tf.cast(self.review_user_adj, dtype=tf.int32)) concate_vecs = tf.concat([ri, self.review_vecs, ru, self.homo_vecs], axis=1) return concate_vecs
class GEMLayer(Layer): "This layer equals to the equation (8) in\n paper 'Heterogeneous Graph Neural Networks for Malicious Account Detection.'\n " def __init__(self, placeholders, nodes, device_num, embedding, encoding, name=None, **kwargs): super(GEMLayer, self).__init__(**kwargs) self.nodes = nodes self.devices_num = device_num self.encoding = encoding self.embedding = embedding self.placeholders = placeholders if (name is not None): name = ('/' + name) else: name = '' with tf.variable_scope(((self.name + name) + '_vars')): self.vars['W'] = glorot([embedding, encoding], name='W') self.vars['V'] = glorot([encoding, encoding], name='V') self.vars['alpha'] = glorot([self.devices_num, 1], name='V') if self.logging: self._log_vars() def _call(self, inputs): h1 = tf.matmul(self.placeholders['x'], self.vars['W']) h2 = [] for d in range(self.devices_num): ahv = tf.matmul(tf.matmul(self.placeholders['a'][d], inputs), self.vars['V']) h2.append(ahv) h2 = tf.concat(h2, 0) h2 = tf.reshape(h2, [self.devices_num, (self.nodes * self.encoding)]) h2 = tf.transpose(h2, [1, 0]) h2 = tf.reshape(tf.matmul(h2, tf.nn.softmax(self.vars['alpha'])), [self.nodes, self.encoding]) h = tf.nn.sigmoid((h1 + h2)) return h
class GAT(Layer): "This layer is adapted from PetarV-/GAT.'\n " def __init__(self, dim, attn_drop, ffd_drop, bias_mat, n_heads, name=None, **kwargs): super(GAT, self).__init__(**kwargs) self.dim = dim self.attn_drop = attn_drop self.ffd_drop = ffd_drop self.bias_mat = bias_mat self.n_heads = n_heads if (name is not None): name = ('/' + name) else: name = '' if self.logging: self._log_vars() def attn_head(self, seq, out_sz, bias_mat, activation, in_drop=0.0, coef_drop=0.0, residual=False): conv1d = tf.layers.conv1d with tf.name_scope('my_attn'): if (in_drop != 0.0): seq = tf.nn.dropout(seq, (1.0 - in_drop)) seq_fts = tf.layers.conv1d(seq, out_sz, 1, use_bias=False) f_1 = tf.layers.conv1d(seq_fts, 1, 1) f_2 = tf.layers.conv1d(seq_fts, 1, 1) logits = (f_1 + tf.transpose(f_2, [0, 2, 1])) coefs = tf.nn.softmax((tf.nn.leaky_relu(logits) + bias_mat)) if (coef_drop != 0.0): coefs = tf.nn.dropout(coefs, (1.0 - coef_drop)) if (in_drop != 0.0): seq_fts = tf.nn.dropout(seq_fts, (1.0 - in_drop)) vals = tf.matmul(coefs, seq_fts) ret = tf.contrib.layers.bias_add(vals) if residual: if (seq.shape[(- 1)] != ret.shape[(- 1)]): ret = (ret + conv1d(seq, ret.shape[(- 1)], 1)) else: ret = (ret + seq) return activation(ret) def inference(self, inputs): out = [] for i in range(self.n_heads): out.append(self.attn_head(inputs, bias_mat=self.bias_mat, out_sz=self.dim, activation=tf.nn.elu, in_drop=self.ffd_drop, coef_drop=self.attn_drop, residual=False)) logits = (tf.add_n(out) / self.n_heads) return logits