ChunML

    train_init_op = iterator.make_initializer(train_data)
    val_init_op = iterator.make_initializer(val_data)

    return next_element, train_init_op, val_init_op

ChunML

def reinitializable_input_fn(filenames, labels, train_val_ratio=0.8):
    num_files = len(filenames)
    num_train_files = int(num_files * train_val_ratio)
    train_filenames = filenames[:num_train_files]
    train_labels = labels[:num_train_files]
    val_filenames = filenames[num_train_files:]
    val_labels = labels[num_train_files:]

    train_data = tf.data.Dataset.from_tensor_slices(
        (train_filenames, train_labels))

ChunML

next_element, train_init_op, val_init_op = reinitializable_input_fn(filenames, labels)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    print('Training...')
    for _ in range(5):
        sess.run(train_init_op)
        imgs, labels = sess.run(next_element)
        print('Image shape:', imgs.shape)

ChunML

def get_data_from_file(train_file, batch_size, seq_size):
    with open(train_file, encoding='utf-8') as f:
        text = f.read()

    text = text.split()

ChunML

    word_counts = Counter(text)
    sorted_vocab = sorted(word_counts, key=word_counts.get, reverse=True)
    int_to_vocab = {k:w for k, w in enumerate(sorted_vocab)}
    vocab_to_int = {w:k for k, w in int_to_vocab.items()}
    n_vocab = len(int_to_vocab)

ChunML

    int_text = [vocab_to_int[w] for w in text]
    num_batches = int(len(int_text) / (seq_size * batch_size))
    in_text = int_text[:num_batches * batch_size * seq_size]

ChunML

    out_text = np.zeros_like(in_text)
    out_text[:-1] = in_text[1:]
    out_text[-1] = in_text[0]
    in_text = np.reshape(in_text, (batch_size, -1))
    out_text = np.reshape(out_text, (batch_size, -1))

    return int_to_vocab, vocab_to_int, n_vocab, in_text, out_text

ChunML

print(in_text[:10, :10])
print(out_text[:10, :10])
'''
in_text:
[[ 412  413  414  415   42  416  417    1  418  419]
 [ 247    1    5  479  144   44   33   70  145   21]
 [  92   37   43   25   72  263    7   18  523   24]
 [   3  590  591  592  593  594    3  595    1   95]
 [  54  650   54   80  182    3  651    6  305   19]
 [ 715  716  717  718    3    0  719    1  720    0]

ChunML

def network(batch_size, seq_size, embedding_size, lstm_size, keep_prob, n_vocab, reuse=False):
    with tf.variable_scope('LSTM', reuse=reuse):
        in_op = tf.placeholder(tf.int32, [None, seq_size])
        out_op = tf.placeholder(tf.int32, [None, seq_size])

ChunML

        embedding = tf.get_variable('embedding_weights', [n_vocab, embedding_size])
        embed = tf.nn.embedding_lookup(embedding, in_op)