dlnd_tv_script_generation


TV Script Generation

In this project, you'll generate your own Simpsons TV scripts using RNNs. You'll be using part of the Simpsons dataset of scripts from 27 seasons. The Neural Network you'll build will generate a new TV script for a scene at Moe's Tavern.

Get the Data

The data is already provided for you. You'll be using a subset of the original dataset. It consists of only the scenes in Moe's Tavern. This doesn't include other versions of the tavern, like "Moe's Cavern", "Flaming Moe's", "Uncle Moe's Family Feed-Bag", etc..


In [2]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

data_dir = './data/simpsons/moes_tavern_lines.txt'
text = helper.load_data(data_dir)
# Ignore notice, since we don't use it for analysing the data
text = text[81:]

Explore the Data

Play around with view_sentence_range to view different parts of the data.


In [3]:
view_sentence_range = (0, 10)

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

print('Dataset Stats')
print('Roughly the number of unique words: {}'.format(len({word: None for word in text.split()})))
scenes = text.split('\n\n')
print('Number of scenes: {}'.format(len(scenes)))
sentence_count_scene = [scene.count('\n') for scene in scenes]
print('Average number of sentences in each scene: {}'.format(np.average(sentence_count_scene)))

sentences = [sentence for scene in scenes for sentence in scene.split('\n')]
print('Number of lines: {}'.format(len(sentences)))
word_count_sentence = [len(sentence.split()) for sentence in sentences]
print('Average number of words in each line: {}'.format(np.average(word_count_sentence)))

print()
print('The sentences {} to {}:'.format(*view_sentence_range))
print('\n'.join(text.split('\n')[view_sentence_range[0]:view_sentence_range[1]]))


Dataset Stats
Roughly the number of unique words: 11492
Number of scenes: 262
Average number of sentences in each scene: 15.248091603053435
Number of lines: 4257
Average number of words in each line: 11.50434578341555

The sentences 0 to 10:
Moe_Szyslak: (INTO PHONE) Moe's Tavern. Where the elite meet to drink.
Bart_Simpson: Eh, yeah, hello, is Mike there? Last name, Rotch.
Moe_Szyslak: (INTO PHONE) Hold on, I'll check. (TO BARFLIES) Mike Rotch. Mike Rotch. Hey, has anybody seen Mike Rotch, lately?
Moe_Szyslak: (INTO PHONE) Listen you little puke. One of these days I'm gonna catch you, and I'm gonna carve my name on your back with an ice pick.
Moe_Szyslak: What's the matter Homer? You're not your normal effervescent self.
Homer_Simpson: I got my problems, Moe. Give me another one.
Moe_Szyslak: Homer, hey, you should not drink to forget your problems.
Barney_Gumble: Yeah, you should only drink to enhance your social skills.


Implement Preprocessing Functions

The first thing to do to any dataset is preprocessing. Implement the following preprocessing functions below:

  • Lookup Table
  • Tokenize Punctuation

Lookup Table

To create a word embedding, you first need to transform the words to ids. In this function, create two dictionaries:

  • Dictionary to go from the words to an id, we'll call vocab_to_int
  • Dictionary to go from the id to word, we'll call int_to_vocab

Return these dictionaries in the following tuple (vocab_to_int, int_to_vocab)


In [4]:
from collections import Counter
import numpy as np
import problem_unittests as tests

def create_lookup_tables(text):
    """
    Create lookup tables for vocabulary
    :param text: The text of tv scripts split into words
    :return: A tuple of dicts (vocab_to_int, int_to_vocab)
    """
    words = sorted(Counter(text), reverse=True)
    vocab_to_int = { word: idx for idx, word in enumerate(words) }
    int_to_vocab = { idx: word for word, idx in vocab_to_int.items()}
    return vocab_to_int, int_to_vocab


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_create_lookup_tables(create_lookup_tables)


Tests Passed

Tokenize Punctuation

We'll be splitting the script into a word array using spaces as delimiters. However, punctuations like periods and exclamation marks make it hard for the neural network to distinguish between the word "bye" and "bye!".

Implement the function token_lookup to return a dict that will be used to tokenize symbols like "!" into "||Exclamation_Mark||". Create a dictionary for the following symbols where the symbol is the key and value is the token:

  • Period ( . )
  • Comma ( , )
  • Quotation Mark ( " )
  • Semicolon ( ; )
  • Exclamation mark ( ! )
  • Question mark ( ? )
  • Left Parentheses ( ( )
  • Right Parentheses ( ) )
  • Dash ( -- )
  • Return ( \n )

This dictionary will be used to token the symbols and add the delimiter (space) around it. This separates the symbols as it's own word, making it easier for the neural network to predict on the next word. Make sure you don't use a token that could be confused as a word. Instead of using the token "dash", try using something like "||dash||".


In [5]:
token_period = "||PERIOD||"
token_comma = "||COMMA||"
token_quotation_mark = "||QUOTATION_MARK||"
token_semicolon = "||SEMICOLON||"
token_exclamation_mark = "||EXCLAMATION_MARK||"
token_question_mark = "||QUESTION_MARK||"
token_left_parenthesis = "||LEFT_PARENTHESIS||"
token_right_parenthesis = "||RIGHT_PARENTHESIS||"
token_dash = "||DASH||"
token_return = "||return||"

def token_lookup():
    """
    Generate a dict to turn punctuation into a token.
    :return: Tokenize dictionary where the key is the punctuation and the value is the token
    """
    return {
        ".": token_period,
        ",": token_comma,
        "\"": token_quotation_mark,
        ";": token_semicolon,
        "!": token_exclamation_mark,
        "?": token_question_mark,
        "(": token_left_parenthesis,
        ")": token_right_parenthesis,
        "--": token_dash,
        "\n": token_return
    }

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_tokenize(token_lookup)


Tests Passed

Preprocess all the data and save it

Running the code cell below will preprocess all the data and save it to file.


In [6]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Preprocess Training, Validation, and Testing Data
helper.preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)

Check Point

This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk.


In [7]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
import numpy as np
import problem_unittests as tests

int_text, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()

Build the Neural Network

You'll build the components necessary to build a RNN by implementing the following functions below:

  • get_inputs
  • get_init_cell
  • get_embed
  • build_rnn
  • build_nn
  • get_batches

Check the Version of TensorFlow and Access to GPU


In [8]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer'
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))


TensorFlow Version: 1.0.0
Default GPU Device: /gpu:0

Input

Implement the get_inputs() function to create TF Placeholders for the Neural Network. It should create the following placeholders:

  • Input text placeholder named "input" using the TF Placeholder name parameter.
  • Targets placeholder
  • Learning Rate placeholder

Return the placeholders in the following tuple (Input, Targets, LearningRate)


In [9]:
def get_inputs():
    """
    Create TF Placeholders for input, targets, and learning rate.
    :return: Tuple (input, targets, learning rate)
    """
    p_input = tf.placeholder(tf.int32, [None, None], name="input")
    p_targets = tf.placeholder(tf.int32, [None, None], name="input")
    p_learning_rate = tf.placeholder(tf.float32, name="learning_rate")
    return (p_input, p_targets, p_learning_rate)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_inputs(get_inputs)


Tests Passed

Build RNN Cell and Initialize

Stack one or more BasicLSTMCells in a MultiRNNCell.

  • The Rnn size should be set using rnn_size
  • Initalize Cell State using the MultiRNNCell's zero_state() function
    • Apply the name "initial_state" to the initial state using tf.identity()

Return the cell and initial state in the following tuple (Cell, InitialState)


In [21]:
# Note: I added layer_count as a default parameter
def get_init_cell(batch_size, rnn_size, layer_count=2):
    """
    Create an RNN Cell and initialize it.
    :param batch_size: Size of batches
    :param rnn_size: Size of RNNs
    :return: Tuple (cell, initialize state)
    """
    basic_lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
    multi_rnn_cell = tf.contrib.rnn.MultiRNNCell([basic_lstm] * layer_count)
    initial_state = tf.identity(multi_rnn_cell.zero_state(batch_size, tf.float32), name="initial_state")
    
    return (multi_rnn_cell, initial_state)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_init_cell(get_init_cell)


Tests Passed

Word Embedding

Apply embedding to input_data using TensorFlow. Return the embedded sequence.


In [22]:
def get_embed(input_data, vocab_size, embed_dim):
    """
    Create embedding for <input_data>.
    :param input_data: TF placeholder for text input.
    :param vocab_size: Number of words in vocabulary.
    :param embed_dim: Number of embedding dimensions
    :return: Embedded input.
    """
    embedding = tf.Variable(tf.random_uniform((vocab_size, embed_dim), -1, 1))
    return tf.nn.embedding_lookup(embedding, input_data)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_embed(get_embed)


Tests Passed

Build RNN

You created a RNN Cell in the get_init_cell() function. Time to use the cell to create a RNN.

Return the outputs and final_state state in the following tuple (Outputs, FinalState)


In [23]:
def build_rnn(cell, inputs):
    """
    Create a RNN using a RNN Cell
    :param cell: RNN Cell
    :param inputs: Input text data
    :return: Tuple (Outputs, Final State)
    """
    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)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_build_rnn(build_rnn)


Tests Passed

Build the Neural Network

Apply the functions you implemented above to:

  • Apply embedding to input_data using your get_embed(input_data, vocab_size, embed_dim) function.
  • Build RNN using cell and your build_rnn(cell, inputs) function.
  • Apply a fully connected layer with a linear activation and vocab_size as the number of outputs.

Return the logits and final state in the following tuple (Logits, FinalState)


In [24]:
def build_nn(cell, rnn_size, input_data, vocab_size, embed_dim):
    """
    Build part of the neural network
    :param cell: RNN cell
    :param rnn_size: Size of rnns
    :param input_data: Input data
    :param vocab_size: Vocabulary size
    :param embed_dim: Number of embedding dimensions
    :return: Tuple (Logits, FinalState)
    """
    embed_layer = get_embed(input_data, vocab_size, rnn_size)
    rnn, final_state = build_rnn(cell, embed_layer)
    fully_connected = tf.layers.dense(rnn, units=vocab_size, activation=None)
    return (fully_connected, final_state)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_build_nn(build_nn)


Tests Passed

Batches

Implement get_batches to create batches of input and targets using int_text. The batches should be a Numpy array with the shape (number of batches, 2, batch size, sequence length). Each batch contains two elements:

  • The first element is a single batch of input with the shape [batch size, sequence length]
  • The second element is a single batch of targets with the shape [batch size, sequence length]

If you can't fill the last batch with enough data, drop the last batch.

For exmple, get_batches([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], 3, 2) would return a Numpy array of the following:

[
  # First Batch
  [
    # Batch of Input
    [[ 1  2], [ 7  8], [13 14]]
    # Batch of targets
    [[ 2  3], [ 8  9], [14 15]]
  ]

  # Second Batch
  [
    # Batch of Input
    [[ 3  4], [ 9 10], [15 16]]
    # Batch of targets
    [[ 4  5], [10 11], [16 17]]
  ]

  # Third Batch
  [
    # Batch of Input
    [[ 5  6], [11 12], [17 18]]
    # Batch of targets
    [[ 6  7], [12 13], [18  1]]
  ]
]

Notice that the last target value in the last batch is the first input value of the first batch. In this case, 1. This is a common technique used when creating sequence batches, although it is rather unintuitive.


In [25]:
def get_batches(int_text, batch_size, seq_length):
    """
    Return batches of input and target
    :param int_text: Text with the words replaced by their ids
    :param batch_size: The size of batch
    :param seq_length: The length of sequence
    :return: Batches as a Numpy array
    """
    
    total_sequences = len(int_text) // seq_length
    
    fixed_ints = int_text[:seq_length * total_sequences]
    
    result = []
    current_batch_input = []
    current_batch_output = []
    read_sequences_count = 0
    for index in range(0, len(fixed_ints), seq_length):
        batch_input = fixed_ints[index : index + seq_length] # take [x, x+1, x+2, ..., x+seq_length-1] -> seq_length elements
        batch_output = fixed_ints[index + 1 : index + seq_length + 1] # take [x+1, x+2, ..., x+seq_length] -> seq_length elements
        
        current_batch_input.append(batch_input)
        current_batch_output.append(batch_output)

        read_sequences_count += 1
        # It is possible we don't complete a batch. In that case, this if won't execute and the result won't be added.
        if read_sequences_count == batch_size:
            result.append([ current_batch_input, current_batch_output ])
            current_batch_input = []
            current_batch_output = []
            read_sequences_count = 0
    
    return np.array(result)

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_batches(get_batches)


Tests Passed

Neural Network Training

Hyperparameters

Tune the following parameters:

  • Set num_epochs to the number of epochs.
  • Set batch_size to the batch size.
  • Set rnn_size to the size of the RNNs.
  • Set embed_dim to the size of the embedding.
  • Set seq_length to the length of sequence.
  • Set learning_rate to the learning rate.
  • Set show_every_n_batches to the number of batches the neural network should print progress.

In [26]:
# Number of Epochs
num_epochs = 500
# Batch Size
batch_size = 512
# RNN Size
rnn_size = 100
# Embedding Dimension Size
embed_dim = None
# Sequence Length
seq_length = 20
# Learning Rate
learning_rate = 0.01
# Show stats for every n number of batches
show_every_n_batches = 10

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
save_dir = './save'

Build the Graph

Build the graph using the neural network you implemented.


In [27]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
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[0], 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[0], input_data_shape[1]]))

    # Optimizer
    optimizer = tf.train.AdamOptimizer(lr)

    # Gradient Clipping
    gradients = optimizer.compute_gradients(cost)
    capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients if grad is not None]
    train_op = optimizer.apply_gradients(capped_gradients)

Train

Train the neural network on the preprocessed data. If you have a hard time getting a good loss, check the forums to see if anyone is having the same problem.


In [28]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
batches = get_batches(int_text, batch_size, seq_length)

with tf.Session(graph=train_graph) as sess:
    sess.run(tf.global_variables_initializer())

    for epoch_i in range(num_epochs):
        state = sess.run(initial_state, {input_text: batches[0][0]})

        for batch_i, (x, y) in enumerate(batches):
            feed = {
                input_text: x,
                targets: y,
                initial_state: state,
                lr: learning_rate}
            train_loss, state, _ = sess.run([cost, final_state, train_op], feed)

            # Show every <show_every_n_batches> batches
            if (epoch_i * len(batches) + batch_i) % show_every_n_batches == 0:
                print('Epoch {:>3} Batch {:>4}/{}   train_loss = {:.3f}'.format(
                    epoch_i,
                    batch_i,
                    len(batches),
                    train_loss))

    # Save Model
    saver = tf.train.Saver()
    saver.save(sess, save_dir)
    print('Model Trained and Saved')


Epoch   0 Batch    0/6   train_loss = 8.822
Epoch   1 Batch    4/6   train_loss = 6.400
Epoch   3 Batch    2/6   train_loss = 5.936
Epoch   5 Batch    0/6   train_loss = 5.880
Epoch   6 Batch    4/6   train_loss = 6.002
Epoch   8 Batch    2/6   train_loss = 5.821
Epoch  10 Batch    0/6   train_loss = 5.760
Epoch  11 Batch    4/6   train_loss = 5.796
Epoch  13 Batch    2/6   train_loss = 5.559
Epoch  15 Batch    0/6   train_loss = 5.507
Epoch  16 Batch    4/6   train_loss = 5.577
Epoch  18 Batch    2/6   train_loss = 5.311
Epoch  20 Batch    0/6   train_loss = 5.254
Epoch  21 Batch    4/6   train_loss = 5.287
Epoch  23 Batch    2/6   train_loss = 5.044
Epoch  25 Batch    0/6   train_loss = 5.051
Epoch  26 Batch    4/6   train_loss = 5.090
Epoch  28 Batch    2/6   train_loss = 4.838
Epoch  30 Batch    0/6   train_loss = 4.837
Epoch  31 Batch    4/6   train_loss = 4.888
Epoch  33 Batch    2/6   train_loss = 4.651
Epoch  35 Batch    0/6   train_loss = 4.670
Epoch  36 Batch    4/6   train_loss = 4.742
Epoch  38 Batch    2/6   train_loss = 4.515
Epoch  40 Batch    0/6   train_loss = 4.552
Epoch  41 Batch    4/6   train_loss = 4.602
Epoch  43 Batch    2/6   train_loss = 4.358
Epoch  45 Batch    0/6   train_loss = 4.359
Epoch  46 Batch    4/6   train_loss = 4.431
Epoch  48 Batch    2/6   train_loss = 4.221
Epoch  50 Batch    0/6   train_loss = 4.239
Epoch  51 Batch    4/6   train_loss = 4.295
Epoch  53 Batch    2/6   train_loss = 4.099
Epoch  55 Batch    0/6   train_loss = 4.100
Epoch  56 Batch    4/6   train_loss = 4.163
Epoch  58 Batch    2/6   train_loss = 3.998
Epoch  60 Batch    0/6   train_loss = 3.958
Epoch  61 Batch    4/6   train_loss = 4.025
Epoch  63 Batch    2/6   train_loss = 3.816
Epoch  65 Batch    0/6   train_loss = 3.837
Epoch  66 Batch    4/6   train_loss = 3.872
Epoch  68 Batch    2/6   train_loss = 3.706
Epoch  70 Batch    0/6   train_loss = 3.734
Epoch  71 Batch    4/6   train_loss = 3.764
Epoch  73 Batch    2/6   train_loss = 3.589
Epoch  75 Batch    0/6   train_loss = 3.598
Epoch  76 Batch    4/6   train_loss = 3.632
Epoch  78 Batch    2/6   train_loss = 3.558
Epoch  80 Batch    0/6   train_loss = 3.512
Epoch  81 Batch    4/6   train_loss = 3.536
Epoch  83 Batch    2/6   train_loss = 3.382
Epoch  85 Batch    0/6   train_loss = 3.414
Epoch  86 Batch    4/6   train_loss = 3.405
Epoch  88 Batch    2/6   train_loss = 3.292
Epoch  90 Batch    0/6   train_loss = 3.301
Epoch  91 Batch    4/6   train_loss = 3.348
Epoch  93 Batch    2/6   train_loss = 3.242
Epoch  95 Batch    0/6   train_loss = 3.206
Epoch  96 Batch    4/6   train_loss = 3.201
Epoch  98 Batch    2/6   train_loss = 3.091
Epoch 100 Batch    0/6   train_loss = 3.096
Epoch 101 Batch    4/6   train_loss = 3.090
Epoch 103 Batch    2/6   train_loss = 3.044
Epoch 105 Batch    0/6   train_loss = 3.069
Epoch 106 Batch    4/6   train_loss = 3.013
Epoch 108 Batch    2/6   train_loss = 2.946
Epoch 110 Batch    0/6   train_loss = 2.925
Epoch 111 Batch    4/6   train_loss = 2.901
Epoch 113 Batch    2/6   train_loss = 2.850
Epoch 115 Batch    0/6   train_loss = 2.825
Epoch 116 Batch    4/6   train_loss = 2.816
Epoch 118 Batch    2/6   train_loss = 2.783
Epoch 120 Batch    0/6   train_loss = 2.753
Epoch 121 Batch    4/6   train_loss = 2.754
Epoch 123 Batch    2/6   train_loss = 2.715
Epoch 125 Batch    0/6   train_loss = 2.691
Epoch 126 Batch    4/6   train_loss = 2.645
Epoch 128 Batch    2/6   train_loss = 2.645
Epoch 130 Batch    0/6   train_loss = 2.571
Epoch 131 Batch    4/6   train_loss = 2.563
Epoch 133 Batch    2/6   train_loss = 2.543
Epoch 135 Batch    0/6   train_loss = 2.536
Epoch 136 Batch    4/6   train_loss = 2.578
Epoch 138 Batch    2/6   train_loss = 2.541
Epoch 140 Batch    0/6   train_loss = 2.460
Epoch 141 Batch    4/6   train_loss = 2.441
Epoch 143 Batch    2/6   train_loss = 2.396
Epoch 145 Batch    0/6   train_loss = 2.361
Epoch 146 Batch    4/6   train_loss = 2.330
Epoch 148 Batch    2/6   train_loss = 2.327
Epoch 150 Batch    0/6   train_loss = 2.323
Epoch 151 Batch    4/6   train_loss = 2.320
Epoch 153 Batch    2/6   train_loss = 2.306
Epoch 155 Batch    0/6   train_loss = 2.328
Epoch 156 Batch    4/6   train_loss = 2.242
Epoch 158 Batch    2/6   train_loss = 2.240
Epoch 160 Batch    0/6   train_loss = 2.196
Epoch 161 Batch    4/6   train_loss = 2.155
Epoch 163 Batch    2/6   train_loss = 2.166
Epoch 165 Batch    0/6   train_loss = 2.187
Epoch 166 Batch    4/6   train_loss = 2.128
Epoch 168 Batch    2/6   train_loss = 2.119
Epoch 170 Batch    0/6   train_loss = 2.087
Epoch 171 Batch    4/6   train_loss = 2.057
Epoch 173 Batch    2/6   train_loss = 2.065
Epoch 175 Batch    0/6   train_loss = 2.091
Epoch 176 Batch    4/6   train_loss = 2.093
Epoch 178 Batch    2/6   train_loss = 2.051
Epoch 180 Batch    0/6   train_loss = 2.050
Epoch 181 Batch    4/6   train_loss = 2.011
Epoch 183 Batch    2/6   train_loss = 1.981
Epoch 185 Batch    0/6   train_loss = 2.045
Epoch 186 Batch    4/6   train_loss = 1.990
Epoch 188 Batch    2/6   train_loss = 1.973
Epoch 190 Batch    0/6   train_loss = 1.914
Epoch 191 Batch    4/6   train_loss = 1.874
Epoch 193 Batch    2/6   train_loss = 1.860
Epoch 195 Batch    0/6   train_loss = 1.846
Epoch 196 Batch    4/6   train_loss = 1.834
Epoch 198 Batch    2/6   train_loss = 1.819
Epoch 200 Batch    0/6   train_loss = 1.815
Epoch 201 Batch    4/6   train_loss = 1.814
Epoch 203 Batch    2/6   train_loss = 1.825
Epoch 205 Batch    0/6   train_loss = 1.813
Epoch 206 Batch    4/6   train_loss = 1.736
Epoch 208 Batch    2/6   train_loss = 1.802
Epoch 210 Batch    0/6   train_loss = 1.711
Epoch 211 Batch    4/6   train_loss = 1.703
Epoch 213 Batch    2/6   train_loss = 1.708
Epoch 215 Batch    0/6   train_loss = 1.679
Epoch 216 Batch    4/6   train_loss = 1.658
Epoch 218 Batch    2/6   train_loss = 1.679
Epoch 220 Batch    0/6   train_loss = 1.622
Epoch 221 Batch    4/6   train_loss = 1.648
Epoch 223 Batch    2/6   train_loss = 1.656
Epoch 225 Batch    0/6   train_loss = 1.620
Epoch 226 Batch    4/6   train_loss = 1.590
Epoch 228 Batch    2/6   train_loss = 1.638
Epoch 230 Batch    0/6   train_loss = 1.562
Epoch 231 Batch    4/6   train_loss = 1.542
Epoch 233 Batch    2/6   train_loss = 1.573
Epoch 235 Batch    0/6   train_loss = 1.528
Epoch 236 Batch    4/6   train_loss = 1.510
Epoch 238 Batch    2/6   train_loss = 1.527
Epoch 240 Batch    0/6   train_loss = 1.498
Epoch 241 Batch    4/6   train_loss = 1.466
Epoch 243 Batch    2/6   train_loss = 1.483
Epoch 245 Batch    0/6   train_loss = 1.461
Epoch 246 Batch    4/6   train_loss = 1.459
Epoch 248 Batch    2/6   train_loss = 1.481
Epoch 250 Batch    0/6   train_loss = 1.442
Epoch 251 Batch    4/6   train_loss = 1.420
Epoch 253 Batch    2/6   train_loss = 1.442
Epoch 255 Batch    0/6   train_loss = 1.423
Epoch 256 Batch    4/6   train_loss = 1.378
Epoch 258 Batch    2/6   train_loss = 1.400
Epoch 260 Batch    0/6   train_loss = 1.369
Epoch 261 Batch    4/6   train_loss = 1.323
Epoch 263 Batch    2/6   train_loss = 1.381
Epoch 265 Batch    0/6   train_loss = 1.323
Epoch 266 Batch    4/6   train_loss = 1.318
Epoch 268 Batch    2/6   train_loss = 1.361
Epoch 270 Batch    0/6   train_loss = 1.307
Epoch 271 Batch    4/6   train_loss = 1.294
Epoch 273 Batch    2/6   train_loss = 1.333
Epoch 275 Batch    0/6   train_loss = 1.317
Epoch 276 Batch    4/6   train_loss = 1.350
Epoch 278 Batch    2/6   train_loss = 1.374
Epoch 280 Batch    0/6   train_loss = 1.289
Epoch 281 Batch    4/6   train_loss = 1.273
Epoch 283 Batch    2/6   train_loss = 1.289
Epoch 285 Batch    0/6   train_loss = 1.255
Epoch 286 Batch    4/6   train_loss = 1.215
Epoch 288 Batch    2/6   train_loss = 1.282
Epoch 290 Batch    0/6   train_loss = 1.242
Epoch 291 Batch    4/6   train_loss = 1.225
Epoch 293 Batch    2/6   train_loss = 1.264
Epoch 295 Batch    0/6   train_loss = 1.225
Epoch 296 Batch    4/6   train_loss = 1.170
Epoch 298 Batch    2/6   train_loss = 1.203
Epoch 300 Batch    0/6   train_loss = 1.148
Epoch 301 Batch    4/6   train_loss = 1.113
Epoch 303 Batch    2/6   train_loss = 1.143
Epoch 305 Batch    0/6   train_loss = 1.118
Epoch 306 Batch    4/6   train_loss = 1.071
Epoch 308 Batch    2/6   train_loss = 1.145
Epoch 310 Batch    0/6   train_loss = 1.103
Epoch 311 Batch    4/6   train_loss = 1.126
Epoch 313 Batch    2/6   train_loss = 1.174
Epoch 315 Batch    0/6   train_loss = 1.150
Epoch 316 Batch    4/6   train_loss = 1.097
Epoch 318 Batch    2/6   train_loss = 1.130
Epoch 320 Batch    0/6   train_loss = 1.107
Epoch 321 Batch    4/6   train_loss = 1.065
Epoch 323 Batch    2/6   train_loss = 1.124
Epoch 325 Batch    0/6   train_loss = 1.099
Epoch 326 Batch    4/6   train_loss = 1.089
Epoch 328 Batch    2/6   train_loss = 1.110
Epoch 330 Batch    0/6   train_loss = 1.068
Epoch 331 Batch    4/6   train_loss = 1.073
Epoch 333 Batch    2/6   train_loss = 1.070
Epoch 335 Batch    0/6   train_loss = 1.064
Epoch 336 Batch    4/6   train_loss = 1.022
Epoch 338 Batch    2/6   train_loss = 1.035
Epoch 340 Batch    0/6   train_loss = 1.006
Epoch 341 Batch    4/6   train_loss = 0.973
Epoch 343 Batch    2/6   train_loss = 0.985
Epoch 345 Batch    0/6   train_loss = 0.987
Epoch 346 Batch    4/6   train_loss = 1.011
Epoch 348 Batch    2/6   train_loss = 1.016
Epoch 350 Batch    0/6   train_loss = 0.997
Epoch 351 Batch    4/6   train_loss = 0.989
Epoch 353 Batch    2/6   train_loss = 1.015
Epoch 355 Batch    0/6   train_loss = 0.970
Epoch 356 Batch    4/6   train_loss = 0.935
Epoch 358 Batch    2/6   train_loss = 0.947
Epoch 360 Batch    0/6   train_loss = 0.909
Epoch 361 Batch    4/6   train_loss = 0.894
Epoch 363 Batch    2/6   train_loss = 0.908
Epoch 365 Batch    0/6   train_loss = 0.898
Epoch 366 Batch    4/6   train_loss = 0.875
Epoch 368 Batch    2/6   train_loss = 0.915
Epoch 370 Batch    0/6   train_loss = 0.942
Epoch 371 Batch    4/6   train_loss = 0.903
Epoch 373 Batch    2/6   train_loss = 0.932
Epoch 375 Batch    0/6   train_loss = 0.922
Epoch 376 Batch    4/6   train_loss = 0.905
Epoch 378 Batch    2/6   train_loss = 0.896
Epoch 380 Batch    0/6   train_loss = 0.891
Epoch 381 Batch    4/6   train_loss = 0.868
Epoch 383 Batch    2/6   train_loss = 0.884
Epoch 385 Batch    0/6   train_loss = 0.861
Epoch 386 Batch    4/6   train_loss = 0.836
Epoch 388 Batch    2/6   train_loss = 0.897
Epoch 390 Batch    0/6   train_loss = 0.863
Epoch 391 Batch    4/6   train_loss = 0.844
Epoch 393 Batch    2/6   train_loss = 0.884
Epoch 395 Batch    0/6   train_loss = 0.894
Epoch 396 Batch    4/6   train_loss = 0.900
Epoch 398 Batch    2/6   train_loss = 0.917
Epoch 400 Batch    0/6   train_loss = 0.897
Epoch 401 Batch    4/6   train_loss = 0.884
Epoch 403 Batch    2/6   train_loss = 0.947
Epoch 405 Batch    0/6   train_loss = 0.899
Epoch 406 Batch    4/6   train_loss = 0.867
Epoch 408 Batch    2/6   train_loss = 0.888
Epoch 410 Batch    0/6   train_loss = 0.860
Epoch 411 Batch    4/6   train_loss = 0.843
Epoch 413 Batch    2/6   train_loss = 0.864
Epoch 415 Batch    0/6   train_loss = 0.936
Epoch 416 Batch    4/6   train_loss = 0.885
Epoch 418 Batch    2/6   train_loss = 0.908
Epoch 420 Batch    0/6   train_loss = 0.850
Epoch 421 Batch    4/6   train_loss = 0.821
Epoch 423 Batch    2/6   train_loss = 0.817
Epoch 425 Batch    0/6   train_loss = 0.785
Epoch 426 Batch    4/6   train_loss = 0.749
Epoch 428 Batch    2/6   train_loss = 0.769
Epoch 430 Batch    0/6   train_loss = 0.751
Epoch 431 Batch    4/6   train_loss = 0.748
Epoch 433 Batch    2/6   train_loss = 0.766
Epoch 435 Batch    0/6   train_loss = 0.761
Epoch 436 Batch    4/6   train_loss = 0.733
Epoch 438 Batch    2/6   train_loss = 0.774
Epoch 440 Batch    0/6   train_loss = 0.730
Epoch 441 Batch    4/6   train_loss = 0.724
Epoch 443 Batch    2/6   train_loss = 0.740
Epoch 445 Batch    0/6   train_loss = 0.703
Epoch 446 Batch    4/6   train_loss = 0.676
Epoch 448 Batch    2/6   train_loss = 0.721
Epoch 450 Batch    0/6   train_loss = 0.701
Epoch 451 Batch    4/6   train_loss = 0.690
Epoch 453 Batch    2/6   train_loss = 0.718
Epoch 455 Batch    0/6   train_loss = 0.760
Epoch 456 Batch    4/6   train_loss = 0.749
Epoch 458 Batch    2/6   train_loss = 0.771
Epoch 460 Batch    0/6   train_loss = 0.746
Epoch 461 Batch    4/6   train_loss = 0.734
Epoch 463 Batch    2/6   train_loss = 0.762
Epoch 465 Batch    0/6   train_loss = 0.697
Epoch 466 Batch    4/6   train_loss = 0.673
Epoch 468 Batch    2/6   train_loss = 0.690
Epoch 470 Batch    0/6   train_loss = 0.649
Epoch 471 Batch    4/6   train_loss = 0.628
Epoch 473 Batch    2/6   train_loss = 0.650
Epoch 475 Batch    0/6   train_loss = 0.634
Epoch 476 Batch    4/6   train_loss = 0.620
Epoch 478 Batch    2/6   train_loss = 0.647
Epoch 480 Batch    0/6   train_loss = 0.653
Epoch 481 Batch    4/6   train_loss = 0.657
Epoch 483 Batch    2/6   train_loss = 0.714
Epoch 485 Batch    0/6   train_loss = 0.774
Epoch 486 Batch    4/6   train_loss = 0.776
Epoch 488 Batch    2/6   train_loss = 0.798
Epoch 490 Batch    0/6   train_loss = 0.936
Epoch 491 Batch    4/6   train_loss = 0.958
Epoch 493 Batch    2/6   train_loss = 0.942
Epoch 495 Batch    0/6   train_loss = 0.889
Epoch 496 Batch    4/6   train_loss = 0.820
Epoch 498 Batch    2/6   train_loss = 0.764
Model Trained and Saved

Save Parameters

Save seq_length and save_dir for generating a new TV script.


In [29]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Save parameters for checkpoint
helper.save_params((seq_length, save_dir))

Checkpoint


In [30]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import tensorflow as tf
import numpy as np
import helper
import problem_unittests as tests

_, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()
seq_length, load_dir = helper.load_params()

Implement Generate Functions

Get Tensors

Get tensors from loaded_graph using the function get_tensor_by_name(). Get the tensors using the following names:

  • "input:0"
  • "initial_state:0"
  • "final_state:0"
  • "probs:0"

Return the tensors in the following tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)


In [31]:
def get_tensors(loaded_graph):
    """
    Get input, initial state, final state, and probabilities tensor from <loaded_graph>
    :param loaded_graph: TensorFlow graph loaded from file
    :return: Tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)
    """
    input_tensor = loaded_graph.get_tensor_by_name("input:0")
    initial_state_tensor = loaded_graph.get_tensor_by_name("initial_state:0")
    final_state_tensor = loaded_graph.get_tensor_by_name("final_state:0")
    probabilities_tensor = loaded_graph.get_tensor_by_name("probs:0")
    return (input_tensor, initial_state_tensor, final_state_tensor, probabilities_tensor)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_tensors(get_tensors)


Tests Passed

Choose Word

Implement the pick_word() function to select the next word using probabilities.


In [32]:
def pick_word(probabilities, int_to_vocab):
    """
    Pick the next word in the generated text
    :param probabilities: Probabilites of the next word
    :param int_to_vocab: Dictionary of word ids as the keys and words as the values
    :return: String of the predicted word
    """
    to_choose_from = list(int_to_vocab.values())
    return np.random.choice(to_choose_from, p=probabilities)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_pick_word(pick_word)


Tests Passed

Generate TV Script

This will generate the TV script for you. Set gen_length to the length of TV script you want to generate.


In [33]:
gen_length = 200
# homer_simpson, moe_szyslak, or Barney_Gumble
prime_word = 'moe_szyslak'

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
loaded_graph = tf.Graph()
with tf.Session(graph=loaded_graph) as sess:
    # Load saved model
    loader = tf.train.import_meta_graph(load_dir + '.meta')
    loader.restore(sess, load_dir)

    # Get Tensors from loaded model
    input_text, initial_state, final_state, probs = get_tensors(loaded_graph)

    # Sentences generation setup
    gen_sentences = [prime_word + ':']
    prev_state = sess.run(initial_state, {input_text: np.array([[1]])})

    # Generate sentences
    for n in range(gen_length):
        # Dynamic Input
        dyn_input = [[vocab_to_int[word] for word in gen_sentences[-seq_length:]]]
        dyn_seq_length = len(dyn_input[0])

        # Get Prediction
        probabilities, prev_state = sess.run(
            [probs, final_state],
            {input_text: dyn_input, initial_state: prev_state})
        
        pred_word = pick_word(probabilities[dyn_seq_length-1], int_to_vocab)

        gen_sentences.append(pred_word)
    
    # Remove tokens
    tv_script = ' '.join(gen_sentences)
    for key, token in token_dict.items():
        ending = ' ' if key in ['\n', '(', '"'] else ''
        tv_script = tv_script.replace(' ' + token.lower(), key)
    tv_script = tv_script.replace('\n ', '\n')
    tv_script = tv_script.replace('( ', '(')
        
    print(tv_script)


moe_szyslak: and i'm gonna even hear much each way.(boozy) if we go scout meeting.
quimby_#2: no, would are you just get any way.
moe_szyslak: and i'm looking here dressed the fat struggling.
moe_szyslak:(sweetly) oh," look at that one wasn't all the time.
carl_carlson: and" moe, i'm thinking between five back!
lenny_leonard: what a publishers bed lessons, i'm not gonna be true yes edna, please(slow, a lot to kiss 'em edna thing did i borrow about a way in here, my eye? well, how lisa, i five grand! if we find it. thank you wanna get over in selma. i'm gone.
lenny_leonard:(loud whisper) i'm sorry.
moe_szyslak: yeah, i need no water.
the embarrassed) kill your passion. it's thirty thousand life, homer? when i don't need this place like they ain't allowed back in traffic. boxing tonight. i used to do something trouble.
marge_simpson: wow, we've been meanin' to michael

The TV Script is Nonsensical

It's ok if the TV script doesn't make any sense. We trained on less than a megabyte of text. In order to get good results, you'll have to use a smaller vocabulary or get more data. Luckly there's more data! As we mentioned in the begging of this project, this is a subset of another dataset. We didn't have you train on all the data, because that would take too long. However, you are free to train your neural network on all the data. After you complete the project, of course.

Submitting This Project

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_tv_script_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.


In [ ]: