lstm_crf_train.py 文件源码

def inference(X, weights, bias, reuse = None, trainMode = True):
    word_vectors = tf.nn.embedding_lookup(WORDS, X)
    # [batch_size, 80, 50]

    length = GetLength(X)
    length_64 = tf.cast(length, tf.int64)
    reuse = None if trainMode else True

    #if trainMode:
    #  word_vectors = tf.nn.dropout(word_vectors, 0.5)
    with tf.variable_scope("rnn_fwbw", reuse = reuse) as scope:
        forward_output, _ = tf.nn.dynamic_rnn(
            tf.contrib.rnn.LSTMCell(FLAGS.num_hidden, reuse = reuse),
            word_vectors,
            dtype = tf.float32,
            sequence_length = length,
            scope = "RNN_forward")
        backward_output_, _ = tf.nn.dynamic_rnn(
            tf.contrib.rnn.LSTMCell(FLAGS.num_hidden, reuse = reuse),
            inputs = tf.reverse_sequence(word_vectors,
                                       length_64,
                                       seq_dim = 1),
            dtype = tf.float32,
            sequence_length = length,
            scope = "RNN_backword")

    backward_output = tf.reverse_sequence(backward_output_,
                                          length_64,
                                          seq_dim = 1)

    output = tf.concat([forward_output, backward_output], 2)
    # [batch_size, 80, 200]
    output = tf.reshape(output, [-1, FLAGS.num_hidden * 2])
    if trainMode:
        output = tf.nn.dropout(output, 0.5)

    matricized_unary_scores = tf.matmul(output, weights) + bias
    # [batch_size, 80, 4]
    unary_scores = tf.reshape(
        matricized_unary_scores,
        [-1, FLAGS.max_sentence_len, FLAGS.num_tags])

    return unary_scores, length