我们从Python开源项目中，提取了以下13个代码示例，用于说明如何使用tensorflow.python.ops.rnn_cell.BasicRNNCell()。
def __call__(self, inputs, state, scope=None): """Most basic RNN: output = new_state = activation(W * input + U * state + B).""" with vs.variable_scope(scope or type(self).__name__): # "BasicRNNCell" output = self._activation(_linear([inputs, state], self._num_units, True, use_fp16=self.use_fp16)) return output, output
def __init__(self, num_units): super(Grid1BasicRNNCell, self).__init__( num_units=num_units, num_dims=1, input_dims=0, output_dims=0, priority_dims=0, tied=False, cell_fn=lambda n, i: rnn_cell.BasicRNNCell(num_units=n, input_size=i))
def __init__(self, num_units, tied=False, non_recurrent_fn=None): super(Grid2BasicRNNCell, 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.BasicRNNCell(num_units=n, input_size=i), non_recurrent_fn=non_recurrent_fn)
def setUp(self): self.rnn_cell = rnn_cell.BasicRNNCell(self.NUM_RNN_CELL_UNITS) self.mock_target_column = MockTargetColumn( num_label_columns=self.NUM_LABEL_COLUMNS) location = tf.contrib.layers.sparse_column_with_keys( 'location', keys=['west_side', 'east_side', 'nyc']) location_onehot = tf.contrib.layers.one_hot_column(location) self.context_feature_columns = [location_onehot] wire_cast = tf.contrib.layers.sparse_column_with_keys( 'wire_cast', ['marlo', 'omar', 'stringer']) wire_cast_embedded = tf.contrib.layers.embedding_column( wire_cast, dimension=8) measurements = tf.contrib.layers.real_valued_column( 'measurements', dimension=2) self.sequence_feature_columns = [measurements, wire_cast_embedded] self.columns_to_tensors = { 'location': tf.SparseTensor( indices=[[0, 0], [1, 0], [2, 0]], values=['west_side', 'west_side', 'nyc'], shape=[3, 1]), 'wire_cast': tf.SparseTensor( indices=[[0, 0, 0], [0, 1, 0], [1, 0, 0], [1, 1, 0], [1, 1, 1], [2, 0, 0]], values=[b'marlo', b'stringer', b'omar', b'stringer', b'marlo', b'marlo'], shape=[3, 2, 2]), 'measurements': tf.random_uniform([3, 2, 2])}
def __init__(self, vocabularySize, config_param): self.vocabularySize = vocabularySize self.config = config_param self._inputX = tf.placeholder(tf.int32, [self.config.batch_size, self.config.sequence_size], "InputsX") self._inputTargetsY = tf.placeholder(tf.int32, [self.config.batch_size, self.config.sequence_size], "InputTargetsY") #Converting Input in an Embedded form with tf.device("/cpu:0"): #Tells Tensorflow what GPU to use specifically embedding = tf.get_variable("embedding", [self.vocabularySize, self.config.embeddingSize]) embeddingLookedUp = tf.nn.embedding_lookup(embedding, self._inputX) inputs = tf.split(1, self.config.sequence_size, embeddingLookedUp) inputTensorsAsList = [tf.squeeze(input_, [1]) for input_ in inputs] #Define Tensor RNN singleRNNCell = rnn_cell.BasicRNNCell(self.config.hidden_size) self.multilayerRNN = rnn_cell.MultiRNNCell([singleRNNCell] * self.config.num_layers) self._initial_state = self.multilayerRNN.zero_state(self.config.batch_size, tf.float32) #Defining Logits hidden_layer_output, last_state = rnn.rnn(self.multilayerRNN, inputTensorsAsList, initial_state=self._initial_state) hidden_layer_output = tf.reshape(tf.concat(1, hidden_layer_output), [-1, self.config.hidden_size]) self._logits = tf.nn.xw_plus_b(hidden_layer_output, tf.get_variable("softmax_w", [self.config.hidden_size, self.vocabularySize]), tf.get_variable("softmax_b", [self.vocabularySize])) self._predictionSoftmax = tf.nn.softmax(self._logits) #Define the loss loss = seq2seq.sequence_loss_by_example([self._logits], [tf.reshape(self._inputTargetsY, [-1])], [tf.ones([self.config.batch_size * self.config.sequence_size])], self.vocabularySize) self._cost = tf.div(tf.reduce_sum(loss), self.config.batch_size) self._final_state = last_state
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))
