Python seq2seq.rnn_decoder方法代码示例

# 需要导入模块: from tensorflow.python.ops import seq2seq [as 别名]
# 或者: from tensorflow.python.ops.seq2seq import rnn_decoder [as 别名]
def generate(self):
        inputs = tf.split(1, self.args.seq_length, tf.nn.embedding_lookup(self.embedding, self.input_data))
        inputs = map(lambda i: tf.nn.l2_normalize(i, 1), [tf.squeeze(input_, [1]) for input_ in inputs])

        def loop(prev, i):
            return prev

        with tf.variable_scope('GEN', reuse=self.has_init_seq2seq) as scope:
            self.has_init_seq2seq = True
            if self.args.num_layers == 1:
                outputs, last_state = seq2seq.rnn_decoder(inputs, [self.initial_state1], self.cell, loop_function=loop, scope=scope)
            elif self.args.num_layers == 2:
                outputs, last_state = seq2seq.rnn_decoder(inputs, [self.initial_state1, self.initial_state2], self.cell, loop_function=loop, scope=scope)
            else:
                raise Exception('Unsupported number of layers. Use 1 or 2 layers for now..')
            outputs = map(lambda o: tf.nn.l2_normalize(o, 1), outputs)
        self.outputs = outputs
        return outputs 

# 需要导入模块: from tensorflow.python.ops import seq2seq [as 别名]
# 或者: from tensorflow.python.ops.seq2seq import rnn_decoder [as 别名]
def discriminate_wv(self, input_data_wv):
        with tf.variable_scope('DISC', reuse=self.has_init_seq2seq) as scope:
            self.has_init_seq2seq = True
            output_wv, states_wv = seq2seq.rnn_decoder(input_data_wv, self.initial_state, self.cell, scope=scope)
            predicted_classes_wv = tf.matmul(output_wv[-1], self.fc_layer)
        return predicted_classes_wv 

# 需要导入模块: from tensorflow.python.ops import seq2seq [as 别名]
# 或者: from tensorflow.python.ops.seq2seq import rnn_decoder [as 别名]
def __init__(self, args, infer=False):
        self.args = args
        if infer:
            args.batch_size = 1
            args.seq_length = 1

        if args.model == 'rnn':
            cell_fn = rnn_cell.BasicRNNCell
        elif args.model == 'gru':
            cell_fn = rnn_cell.GRUCell
        elif args.model == 'lstm':
            cell_fn = rnn_cell.BasicLSTMCell
        else:
            raise Exception("model type not supported: {}".format(args.model))

        cell = cell_fn(args.rnn_size)

        self.cell = cell = rnn_cell.MultiRNNCell([cell] * args.num_layers)

        self.input_data = tf.placeholder(tf.int32, [args.batch_size, args.seq_length])
        self.targets = tf.placeholder(tf.int32, [args.batch_size, args.seq_length])
        self.initial_state = cell.zero_state(args.batch_size, tf.float32)

        with tf.variable_scope('rnnlm'):
            softmax_w = tf.get_variable("softmax_w", [args.rnn_size, args.vocab_size])
            softmax_b = tf.get_variable("softmax_b", [args.vocab_size])
            with tf.device("/cpu:0"):
                embedding = tf.get_variable("embedding", [args.vocab_size, args.rnn_size])
                inputs = tf.split(1, args.seq_length, tf.nn.embedding_lookup(embedding, self.input_data))
                inputs = [tf.squeeze(input_, [1]) for input_ in inputs]

        def loop(prev, _):
            prev = tf.matmul(prev, softmax_w) + softmax_b
            prev_symbol = tf.stop_gradient(tf.argmax(prev, 1))
            return tf.nn.embedding_lookup(embedding, prev_symbol)

        outputs, last_state = seq2seq.rnn_decoder(inputs, self.initial_state, cell, loop_function=loop if infer else None, scope='rnnlm')
        output = tf.reshape(tf.concat(1, outputs), [-1, args.rnn_size])
        self.logits = tf.matmul(output, softmax_w) + softmax_b
        self.probs = tf.nn.softmax(self.logits)
        loss = seq2seq.sequence_loss_by_example([self.logits],
                [tf.reshape(self.targets, [-1])],
                [tf.ones([args.batch_size * args.seq_length])],
                args.vocab_size)
        self.cost = tf.reduce_sum(loss) / args.batch_size / args.seq_length
        self.final_state = last_state
        self.lr = tf.Variable(0.0, trainable=False)
        tvars = tf.trainable_variables()
        grads, _ = tf.clip_by_global_norm(tf.gradients(self.cost, tvars),
                args.grad_clip)
        optimizer = tf.train.AdamOptimizer(self.lr)
        self.train_op = optimizer.apply_gradients(zip(grads, tvars)) 

# 需要导入模块: from tensorflow.python.ops import seq2seq [as 别名]
# 或者: from tensorflow.python.ops.seq2seq import rnn_decoder [as 别名]
def __init__(self, args, infer=False):
        self.args = args
        if infer:
            args.batch_size = 1
            args.seq_length = 1

        if args.rnncell == 'rnn':
            cell_fn = rnn_cell.BasicRNNCell
        elif args.rnncell == 'gru':
            cell_fn = rnn_cell.GRUCell
        elif args.rnncell == 'lstm':
            cell_fn = rnn_cell.BasicLSTMCell
        else:
            raise Exception("rnncell type not supported: {}".format(args.rnncell))

        cell = cell_fn(args.rnn_size)
        self.cell = rnn_cell.MultiRNNCell([cell] * args.num_layers)
        self.input_data = tf.placeholder(tf.int32, [args.batch_size, args.seq_length])
        self.targets = tf.placeholder(tf.int32, [args.batch_size, args.seq_length])
        self.initial_state = self.cell.zero_state(args.batch_size, tf.float32)
        with tf.variable_scope('rnnlm'):
            softmax_w = build_weight([args.rnn_size, args.vocab_size],name='soft_w')
            softmax_b = build_weight([args.vocab_size],name='soft_b')
            word_embedding = build_weight([args.vocab_size, args.embedding_size],name='word_embedding')
            inputs_list = tf.split(1, args.seq_length, tf.nn.embedding_lookup(word_embedding, self.input_data))
            inputs_list = [tf.squeeze(input_, [1]) for input_ in inputs_list]
        def loop(prev, _):
            prev = tf.matmul(prev, softmax_w) + softmax_b
            prev_symbol = tf.stop_gradient(tf.argmax(prev, 1))
            return tf.nn.embedding_lookup(embedding, prev_symbol)

	if not args.attention:
            outputs, last_state = seq2seq.rnn_decoder(inputs_list, self.initial_state, self.cell, loop_function=loop if infer else None, scope='rnnlm')
	else:
	    self.attn_length = 5
	    self.attn_size = 32
	    self.attention_states = build_weight([args.batch_size, self.attn_length, self.attn_size]) 
            outputs, last_state = seq2seq.attention_decoder(inputs_list, self.initial_state, self.attention_states, self.cell, loop_function=loop if infer else None, scope='rnnlm')

        self.final_state = last_state
        output = tf.reshape(tf.concat(1, outputs), [-1, args.rnn_size])
        self.logits = tf.matmul(output, softmax_w) + softmax_b
        self.probs = tf.nn.softmax(self.logits)
        loss = seq2seq.sequence_loss_by_example([self.logits],
                [tf.reshape(self.targets, [-1])],
                [tf.ones([args.batch_size * args.seq_length])],
                args.vocab_size)
	# average loss for each word of each timestep
        self.cost = tf.reduce_sum(loss) / args.batch_size / args.seq_length
        self.lr = tf.Variable(0.0, trainable=False)
	self.var_trainable_op = tf.trainable_variables()
        grads, _ = tf.clip_by_global_norm(tf.gradients(self.cost, self.var_trainable_op),
                args.grad_clip)
        optimizer = tf.train.AdamOptimizer(self.lr)
        self.train_op = optimizer.apply_gradients(zip(grads, self.var_trainable_op))
	self.initial_op = tf.initialize_all_variables()
	self.saver = tf.train.Saver(tf.all_variables(),max_to_keep=5,keep_checkpoint_every_n_hours=1)
	self.logfile = args.log_dir+str(datetime.datetime.strftime(datetime.datetime.now(),'%Y-%m-%d %H:%M:%S')+'.txt').replace(' ','').replace('/','')
	self.var_op = tf.all_variables() 

# 需要导入模块: from tensorflow.python.ops import seq2seq [as 别名]
# 或者: from tensorflow.python.ops.seq2seq import rnn_decoder [as 别名]
def __init__(self, args, embedding):
        self.args = args

        if args.model == 'rnn':
            cell_fn = rnn_cell.BasicRNNCell
        elif args.model == 'gru':
            cell_fn = rnn_cell.GRUCell
        elif args.model == 'lstm':
            cell_fn = rnn_cell.BasicLSTMCell
        else:
            raise Exception("model type not supported: {}".format(args.model))

        cell = cell_fn(args.rnn_size)

        self.cell = cell = rnn_cell.MultiRNNCell([cell] * args.num_layers)

        self.input_data = tf.placeholder(tf.int32, [args.batch_size, args.seq_length], name='STAND_input')
        self.targets = tf.placeholder(tf.int32, [args.batch_size, args.seq_length], name='STAND_targets')
        self.initial_state = cell.zero_state(args.batch_size, tf.float32)
        self.embedding = embedding
        with tf.variable_scope('STAND'):
            softmax_w = tf.get_variable("softmax_w", [args.rnn_size, args.vocab_size])
            softmax_b = tf.get_variable("softmax_b", [args.vocab_size])
            inputs = tf.split(1, args.seq_length, tf.nn.embedding_lookup(self.embedding, self.input_data))
            inputs = map(lambda i: tf.nn.l2_normalize(i, 1), [tf.squeeze(input_, [1]) for input_ in inputs])

        def loop(prev, i):
            prev = tf.matmul(prev, softmax_w) + softmax_b
            prev_symbol = tf.stop_gradient(tf.argmax(prev, 1))
            return tf.nn.l2_normalize(tf.nn.embedding_lookup(embedding, prev_symbol), 1)

        o, _ = seq2seq.rnn_decoder(inputs, self.initial_state, cell, loop_function=None, scope='STAND')
        with tf.variable_scope('STAND', reuse=True) as scope:
            sf_o, _ = seq2seq.rnn_decoder(inputs, self.initial_state, cell, loop_function=loop, scope=scope)
        output = tf.reshape(tf.concat(1, o), [-1, args.rnn_size])
        self.logits = tf.matmul(output, softmax_w) + softmax_b
        self.probs = tf.nn.softmax(self.logits)

        sf_output = tf.reshape(tf.concat(1, sf_o), [-1, args.rnn_size])
        self_feed_logits = tf.matmul(sf_output, softmax_w) + softmax_b
        self.self_feed_probs = tf.nn.softmax(self_feed_logits)

        loss = seq2seq.sequence_loss_by_example([self.logits],
                [tf.reshape(self.targets, [-1])],
                [tf.ones([args.batch_size * args.seq_length])],
                args.vocab_size)
        self.loss = tf.reduce_sum(loss) / args.batch_size / args.seq_length
        self.lr = tf.Variable(0.0, trainable=False)
        tvars = tf.trainable_variables()
        grads, _ = tf.clip_by_global_norm(tf.gradients(self.loss, tvars),
                args.grad_clip)
        for g, v in zip(grads, tvars):
            print v.name
        optimizer = tf.train.AdamOptimizer(self.lr)
        self.train_op = optimizer.apply_gradients(zip(grads, tvars))