#自動寫劇本#深度網絡結構

• 初始化網絡

def get_init_cell(batch_size, rnn_size):
    lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
    cell = tf.contrib.rnn.MultiRNNCell([lstm] * )
    initial_state = cell.zero_state(batch_size, tf.float32)
    initial_state = tf.identity(initial_state, name='initial_state') 
    return (cell, initial_state)
           

• 輸入

def get_embed(input_data, vocab_size, embed_dim):
    embedding = tf.Variable(tf.random_uniform((vocab_size, embed_dim), -, ))
    embed = tf.nn.embedding_lookup(embedding, input_data)
    return embed
           

• 單個rnn節點

def build_rnn(cell, inputs):
    outputs, final_state = tf.nn.dynamic_rnn(cell, inputs, dtype=tf.float32)
    final_state = tf.identity(final_state, name='final_state')
    return (outputs, final_state)
           

• 整個rnn網絡

def build_nn(cell, rnn_size, input_data, vocab_size, embed_dim):
    embedded_input = get_embed(input_data, vocab_size, embed_dim)
    outputs, final_state = build_rnn(cell, embedded_input)
    print final_state.shape
    logits = tf.contrib.layers.fully_connected(outputs, vocab_size)
    return (logits, final_state)
           

• build the graph

from tensorflow.contrib import seq2seq
train_graph = tf.Graph()
with train_graph.as_default():
    vocab_size = len(int_to_vocab)
    input_text, targets, lr = get_inputs()
    input_data_shape = tf.shape(input_text)
    cell, initial_state = get_init_cell(input_data_shape[], rnn_size)
        logits, final_state = build_nn(cell, rnn_size, input_text, vocab_size, embed_dim)
    # Probabilities for generating words
    probs = tf.nn.softmax(logits, name='probs')
    # Loss function
    cost = seq2seq.sequence_loss(
        logits,
        targets,
        tf.ones([input_data_shape[], input_data_shape[]]))
    # Optimizer
    optimizer = tf.train.AdamOptimizer(lr)
    # Gradient Clipping
    gradients = optimizer.compute_gradients(cost)
    capped_gradients = [(tf.clip_by_value(grad, -, ), var) for grad, var in gradients if grad is not None]
    train_op = optimizer.apply_gradients(capped_gradients)
           

Ending