我们从Python开源项目中，提取了以下9个代码示例，用于说明如何使用tensorflow.python.ops.rnn_cell.GRUCell()。
def __init__(self, num_units, tied=False, non_recurrent_fn=None): super(Grid2GRUCell, self).__init__( num_units=num_units, num_dims=2, input_dims=0, output_dims=0, priority_dims=0, tied=tied, non_recurrent_dims=None if non_recurrent_fn is None else 0, cell_fn=lambda n, i: rnn_cell.GRUCell(num_units=n, input_size=i), non_recurrent_fn=non_recurrent_fn)
def __init__(self, args): self.args = args if args.disc_model == 'rnn': cell_fn = rnn_cell.BasicRNNCell elif args.disc_model == 'gru': cell_fn = rnn_cell.GRUCell elif args.disc_model == 'lstm': cell_fn = rnn_cell.BasicLSTMCell else: raise Exception("model type not supported: {}".format(args.model)) self.embedding = tf.Variable(tf.random_uniform([self.args.vocab_size, self.args.rnn_size], minval=-.05, maxval=.05, dtype=tf.float32), name='embedding') with tf.variable_scope('DISC') as scope: cell = cell_fn(args.rnn_size) self.cell = cell = rnn_cell.MultiRNNCell([cell] * args.num_layers) # If the input data is given as word tokens, feed this value self.input_data_text = tf.placeholder(tf.int32, [args.batch_size, args.seq_length], name='input_data_text') #self.input_data_text = tf.Variable(tf.zeros((args.batch_size, args.seq_length), dtype=tf.int32), name='input_data_text') self.initial_state = cell.zero_state(args.batch_size, tf.float32) # Fully connected layer is applied to the final state to determine the output class self.fc_layer = tf.Variable(tf.random_normal([args.rnn_size, 1], stddev=0.35, dtype=tf.float32), name='disc_fc_layer') self.lr = tf.Variable(0.0, trainable=False, name='learning_rate') self.has_init_seq2seq = False
def __init__(self, args): self.args = args if args.gen_model == 'rnn': cell_fn = rnn_cell.BasicRNNCell elif args.gen_model == 'gru': cell_fn = rnn_cell.GRUCell elif args.gen_model == 'lstm': cell_fn = rnn_cell.BasicLSTMCell else: raise Exception("model type not supported: {}".format(args.model)) with tf.variable_scope('GEN') as scope: cell = cell_fn(args.rnn_size) self.cell = cell = rnn_cell.MultiRNNCell([cell] * args.num_layers) # sequence of word tokens taken as input self.input_data = tf.placeholder(tf.int32, [args.batch_size, args.seq_length], name='input_data') self.latent_state = tf.placeholder(tf.float32, [args.batch_size, args.latent_size]) # weights to map the latent state into the (usually) bigger initial state # right now this only works for rnn (other more complex models have more than # one initial state which needs to be given a value) # Right now we support up to two layers (state1 and state2) self.latent_to_initial_state1 = tf.Variable(tf.random_normal([args.latent_size, args.rnn_size], stddev=0.35, dtype=tf.float32), name='latent_to_intial_state1') self.latent_to_initial_state2 = tf.Variable(tf.random_normal([args.latent_size, args.rnn_size], stddev=0.35, dtype=tf.float32), name='latent_to_intial_state2') self.initial_state1 = tf.matmul(self.latent_state, self.latent_to_initial_state1) self.initial_state2 = tf.matmul(self.latent_state, self.latent_to_initial_state2) # these are the actual approximate word vectors generated by the model self.outputs = tf.placeholder(tf.float32, [args.seq_length, args.batch_size, args.rnn_size]) self.lr = tf.Variable(0.0, trainable=False, name='learning_rate') self.has_init_seq2seq = False
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))
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()
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))
def decode_model_with_buckets(encoder_inputs, decoder_inputs, targets, weights, buckets, seq2seq, softmax_loss_function=None, per_example_loss=False, name=None): """Create a sequence-to-sequence models with support for bucketing. The seq2seq argument is a function that defines a sequence-to-sequence models, e.g., seq2seq = lambda x, y: basic_rnn_seq2seq(x, y, rnn_cell.GRUCell(24)) Args: encoder_inputs: A list of Tensors to feed the encoder; first seq2seq input. decoder_inputs: A list of Tensors to feed the decoder; second seq2seq input. targets: A list of 1D batch-sized int32 Tensors (desired output sequence). weights: List of 1D batch-sized float-Tensors to weight the targets. buckets: A list of pairs of (input size, output size) for each bucket. seq2seq: A sequence-to-sequence models function; it takes 2 input that agree with encoder_inputs and decoder_inputs, and returns a pair consisting of outputs and states (as, e.g., basic_rnn_seq2seq). softmax_loss_function: Function (inputs-batch, labels-batch) -> loss-batch to be used instead of the standard softmax (the default if this is None). per_example_loss: Boolean. If set, the returned loss will be a batch-sized tensor of losses for each sequence in the batch. If unset, it will be a scalar with the averaged loss from all examples. name: Optional name for this operation, defaults to "model_with_buckets". Returns: A tuple of the form (outputs, losses), where: outputs: The outputs for each bucket. Its j'th element consists of a list of 2D Tensors of shape [batch_size x num_decoder_symbols] (jth outputs). losses: List of scalar Tensors, representing losses for each bucket, or, if per_example_loss is set, a list of 1D batch-sized float Tensors. Raises: ValueError: If length of encoder_inputsut, targets, or weights is smaller than the largest (last) bucket. """ if len(encoder_inputs) < buckets[-1][0]: raise ValueError("Length of encoder_inputs (%d) must be at least that of la" "st bucket (%d)." % (len(encoder_inputs), buckets[-1][0])) if len(targets) < buckets[-1][1]: raise ValueError("Length of targets (%d) must be at least that of last" "bucket (%d)." % (len(targets), buckets[-1][1])) if len(weights) < buckets[-1][1]: raise ValueError("Length of weights (%d) must be at least that of last" "bucket (%d)." % (len(weights), buckets[-1][1])) all_inputs = encoder_inputs + decoder_inputs + targets + weights states = [] outputs = [] with ops.name_scope(name, "model_with_buckets", all_inputs): for j, bucket in enumerate(buckets): with variable_scope.variable_scope(variable_scope.get_variable_scope(), reuse=True if j > 0 else None): bucket_outputs, bucket_states = seq2seq(encoder_inputs[:bucket[0]], decoder_inputs[:bucket[1]]) states.append(bucket_states) outputs.append(bucket_outputs) return outputs, states
