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 [1]:
"""
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 [2]:
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 [3]:
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)
    """
    # TODO: Implement Function
    words = list(set(text))
    vocab_to_int = {}
    int_to_vocab = {}
    for i, word in enumerate(words):
        vocab_to_int[word] = i
        int_to_vocab[i] = word
    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 [4]:
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
    """
    # TODO: Implement Function
    puncs = {".": "Period", ",": "Comma", "\"": "QuotationMark", 
             ";": "Semicolon", "!": "ExclamationMark", "?": "QuestionMark", 
             "(": "LeftParantheses", ")": "RightParantheses", "--": "Dash", "\n": "Return"
            }
    return {key: "||{}||".format(val) for key, val in puncs.items()}

"""
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 [5]:
"""
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 [6]:
"""
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 [7]:
"""
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: 1.0.1
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 [8]:
def get_inputs():
    """
    Create TF Placeholders for input, targets, and learning rate.
    :return: Tuple (input, targets, learning rate)
    """
    # TODO: Implement Function
    inputs = tf.placeholder(tf.int32, [None, None], name="input")
    targets = tf.placeholder(tf.int32, [None, None], name="targets")
    lr = tf.placeholder(tf.float32, name="learning_rate")
    return inputs, targets, lr


"""
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 [9]:
lstm_layers = 2

def get_init_cell(batch_size, rnn_size):
    """
    Create an RNN Cell and initialize it.
    :param batch_size: Size of batches
    :param rnn_size: Size of RNNs
    :return: Tuple (cell, initialize state)
    """
    
    lstm = tf.contrib.rnn.BasicLSTMCell(rnn_size)
        
    # Stack up multiple LSTM layers, for deep learning
    cell = tf.contrib.rnn.MultiRNNCell([lstm] * lstm_layers)
    
    # Getting an initial state of all zeros
    initial_state = tf.identity(cell.zero_state(batch_size, tf.float32), name="initial_state")

    return 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 [10]:
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.
    """
    # TODO: Implement Function
    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 [11]:
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)
    """
    # TODO: Implement Function
    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 [12]:
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)
    """
    # TODO: Implement Function
    inputs = get_embed(input_data, vocab_size, rnn_size)
    outputs, final_state = build_rnn(cell, inputs)
    predictions = tf.contrib.layers.fully_connected(outputs, vocab_size, activation_fn=None)
    return predictions, 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 [13]:
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
    """
    # TODO: Implement Function
    n_batches = (len(int_text) - 1) // (batch_size * seq_length)
    batches = np.zeros((n_batches, 2, batch_size, seq_length))
    x = int_text[:]
    y = int_text[1:]
    for i in range(batch_size):
        for j in range(n_batches):
            k = (i * n_batches + j) * seq_length
            batches[j, 0, i] = x[k: k + seq_length]
            batches[j, 1, i] = y[k: k + seq_length]
    return batches

# Self check
print(get_batches([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], 2, 3))
  
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_get_batches(get_batches)


[[[[  1.   2.   3.]
   [  7.   8.   9.]]

  [[  2.   3.   4.]
   [  8.   9.  10.]]]


 [[[  4.   5.   6.]
   [ 10.  11.  12.]]

  [[  5.   6.   7.]
   [ 11.  12.  13.]]]]
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 [14]:
# Number of Epochs
num_epochs = 200

# Batch Size
batch_size = 128

# RNN Size
rnn_size = 2000

# Sequence Length
seq_length = 50

# Learning Rate
learning_rate = 0.0002

# Show stats for every n number of batches
show_every_n_batches = 5

"""
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 [15]:
"""
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 [16]:
"""
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    0/10   train_loss = 8.822
Epoch   0 Batch    5/10   train_loss = 6.949
Epoch   1 Batch    0/10   train_loss = 6.208
Epoch   1 Batch    5/10   train_loss = 6.051
Epoch   2 Batch    0/10   train_loss = 6.028
Epoch   2 Batch    5/10   train_loss = 5.871
Epoch   3 Batch    0/10   train_loss = 5.851
Epoch   3 Batch    5/10   train_loss = 5.733
Epoch   4 Batch    0/10   train_loss = 5.700
Epoch   4 Batch    5/10   train_loss = 5.593
Epoch   5 Batch    0/10   train_loss = 5.569
Epoch   5 Batch    5/10   train_loss = 5.463
Epoch   6 Batch    0/10   train_loss = 5.439
Epoch   6 Batch    5/10   train_loss = 5.347
Epoch   7 Batch    0/10   train_loss = 5.329
Epoch   7 Batch    5/10   train_loss = 5.247
Epoch   8 Batch    0/10   train_loss = 5.232
Epoch   8 Batch    5/10   train_loss = 5.157
Epoch   9 Batch    0/10   train_loss = 5.143
Epoch   9 Batch    5/10   train_loss = 5.071
Epoch  10 Batch    0/10   train_loss = 5.057
Epoch  10 Batch    5/10   train_loss = 4.990
Epoch  11 Batch    0/10   train_loss = 4.977
Epoch  11 Batch    5/10   train_loss = 4.913
Epoch  12 Batch    0/10   train_loss = 4.899
Epoch  12 Batch    5/10   train_loss = 4.841
Epoch  13 Batch    0/10   train_loss = 4.828
Epoch  13 Batch    5/10   train_loss = 4.776
Epoch  14 Batch    0/10   train_loss = 4.756
Epoch  14 Batch    5/10   train_loss = 4.707
Epoch  15 Batch    0/10   train_loss = 4.696
Epoch  15 Batch    5/10   train_loss = 4.643
Epoch  16 Batch    0/10   train_loss = 4.627
Epoch  16 Batch    5/10   train_loss = 4.582
Epoch  17 Batch    0/10   train_loss = 4.565
Epoch  17 Batch    5/10   train_loss = 4.517
Epoch  18 Batch    0/10   train_loss = 4.501
Epoch  18 Batch    5/10   train_loss = 4.467
Epoch  19 Batch    0/10   train_loss = 4.445
Epoch  19 Batch    5/10   train_loss = 4.398
Epoch  20 Batch    0/10   train_loss = 4.395
Epoch  20 Batch    5/10   train_loss = 4.364
Epoch  21 Batch    0/10   train_loss = 4.369
Epoch  21 Batch    5/10   train_loss = 4.335
Epoch  22 Batch    0/10   train_loss = 4.293
Epoch  22 Batch    5/10   train_loss = 4.253
Epoch  23 Batch    0/10   train_loss = 4.241
Epoch  23 Batch    5/10   train_loss = 4.206
Epoch  24 Batch    0/10   train_loss = 4.187
Epoch  24 Batch    5/10   train_loss = 4.148
Epoch  25 Batch    0/10   train_loss = 4.128
Epoch  25 Batch    5/10   train_loss = 4.099
Epoch  26 Batch    0/10   train_loss = 4.089
Epoch  26 Batch    5/10   train_loss = 4.069
Epoch  27 Batch    0/10   train_loss = 4.058
Epoch  27 Batch    5/10   train_loss = 4.060
Epoch  28 Batch    0/10   train_loss = 4.011
Epoch  28 Batch    5/10   train_loss = 3.974
Epoch  29 Batch    0/10   train_loss = 3.954
Epoch  29 Batch    5/10   train_loss = 3.914
Epoch  30 Batch    0/10   train_loss = 3.908
Epoch  30 Batch    5/10   train_loss = 3.890
Epoch  31 Batch    0/10   train_loss = 3.868
Epoch  31 Batch    5/10   train_loss = 3.858
Epoch  32 Batch    0/10   train_loss = 3.856
Epoch  32 Batch    5/10   train_loss = 3.845
Epoch  33 Batch    0/10   train_loss = 3.867
Epoch  33 Batch    5/10   train_loss = 3.783
Epoch  34 Batch    0/10   train_loss = 3.767
Epoch  34 Batch    5/10   train_loss = 3.727
Epoch  35 Batch    0/10   train_loss = 3.704
Epoch  35 Batch    5/10   train_loss = 3.684
Epoch  36 Batch    0/10   train_loss = 3.657
Epoch  36 Batch    5/10   train_loss = 3.639
Epoch  37 Batch    0/10   train_loss = 3.610
Epoch  37 Batch    5/10   train_loss = 3.607
Epoch  38 Batch    0/10   train_loss = 3.582
Epoch  38 Batch    5/10   train_loss = 3.560
Epoch  39 Batch    0/10   train_loss = 3.551
Epoch  39 Batch    5/10   train_loss = 3.551
Epoch  40 Batch    0/10   train_loss = 3.530
Epoch  40 Batch    5/10   train_loss = 3.474
Epoch  41 Batch    0/10   train_loss = 3.427
Epoch  41 Batch    5/10   train_loss = 3.408
Epoch  42 Batch    0/10   train_loss = 3.403
Epoch  42 Batch    5/10   train_loss = 3.382
Epoch  43 Batch    0/10   train_loss = 3.338
Epoch  43 Batch    5/10   train_loss = 3.333
Epoch  44 Batch    0/10   train_loss = 3.315
Epoch  44 Batch    5/10   train_loss = 3.262
Epoch  45 Batch    0/10   train_loss = 3.239
Epoch  45 Batch    5/10   train_loss = 3.221
Epoch  46 Batch    0/10   train_loss = 3.216
Epoch  46 Batch    5/10   train_loss = 3.242
Epoch  47 Batch    0/10   train_loss = 3.169
Epoch  47 Batch    5/10   train_loss = 3.163
Epoch  48 Batch    0/10   train_loss = 3.115
Epoch  48 Batch    5/10   train_loss = 3.100
Epoch  49 Batch    0/10   train_loss = 3.076
Epoch  49 Batch    5/10   train_loss = 3.062
Epoch  50 Batch    0/10   train_loss = 3.030
Epoch  50 Batch    5/10   train_loss = 3.020
Epoch  51 Batch    0/10   train_loss = 2.981
Epoch  51 Batch    5/10   train_loss = 2.952
Epoch  52 Batch    0/10   train_loss = 2.928
Epoch  52 Batch    5/10   train_loss = 2.929
Epoch  53 Batch    0/10   train_loss = 2.930
Epoch  53 Batch    5/10   train_loss = 2.914
Epoch  54 Batch    0/10   train_loss = 2.871
Epoch  54 Batch    5/10   train_loss = 2.931
Epoch  55 Batch    0/10   train_loss = 2.838
Epoch  55 Batch    5/10   train_loss = 2.879
Epoch  56 Batch    0/10   train_loss = 2.880
Epoch  56 Batch    5/10   train_loss = 2.879
Epoch  57 Batch    0/10   train_loss = 2.843
Epoch  57 Batch    5/10   train_loss = 2.852
Epoch  58 Batch    0/10   train_loss = 2.809
Epoch  58 Batch    5/10   train_loss = 2.810
Epoch  59 Batch    0/10   train_loss = 2.798
Epoch  59 Batch    5/10   train_loss = 2.785
Epoch  60 Batch    0/10   train_loss = 2.765
Epoch  60 Batch    5/10   train_loss = 2.770
Epoch  61 Batch    0/10   train_loss = 2.700
Epoch  61 Batch    5/10   train_loss = 2.695
Epoch  62 Batch    0/10   train_loss = 2.694
Epoch  62 Batch    5/10   train_loss = 2.707
Epoch  63 Batch    0/10   train_loss = 2.672
Epoch  63 Batch    5/10   train_loss = 2.629
Epoch  64 Batch    0/10   train_loss = 2.561
Epoch  64 Batch    5/10   train_loss = 2.550
Epoch  65 Batch    0/10   train_loss = 2.515
Epoch  65 Batch    5/10   train_loss = 2.540
Epoch  66 Batch    0/10   train_loss = 2.531
Epoch  66 Batch    5/10   train_loss = 2.647
Epoch  67 Batch    0/10   train_loss = 2.599
Epoch  67 Batch    5/10   train_loss = 2.517
Epoch  68 Batch    0/10   train_loss = 2.471
Epoch  68 Batch    5/10   train_loss = 2.450
Epoch  69 Batch    0/10   train_loss = 2.374
Epoch  69 Batch    5/10   train_loss = 2.361
Epoch  70 Batch    0/10   train_loss = 2.329
Epoch  70 Batch    5/10   train_loss = 2.319
Epoch  71 Batch    0/10   train_loss = 2.271
Epoch  71 Batch    5/10   train_loss = 2.270
Epoch  72 Batch    0/10   train_loss = 2.258
Epoch  72 Batch    5/10   train_loss = 2.326
Epoch  73 Batch    0/10   train_loss = 2.258
Epoch  73 Batch    5/10   train_loss = 2.240
Epoch  74 Batch    0/10   train_loss = 2.249
Epoch  74 Batch    5/10   train_loss = 2.285
Epoch  75 Batch    0/10   train_loss = 2.262
Epoch  75 Batch    5/10   train_loss = 2.380
Epoch  76 Batch    0/10   train_loss = 2.241
Epoch  76 Batch    5/10   train_loss = 2.170
Epoch  77 Batch    0/10   train_loss = 2.109
Epoch  77 Batch    5/10   train_loss = 2.109
Epoch  78 Batch    0/10   train_loss = 2.044
Epoch  78 Batch    5/10   train_loss = 2.029
Epoch  79 Batch    0/10   train_loss = 1.957
Epoch  79 Batch    5/10   train_loss = 1.969
Epoch  80 Batch    0/10   train_loss = 1.906
Epoch  80 Batch    5/10   train_loss = 1.932
Epoch  81 Batch    0/10   train_loss = 1.884
Epoch  81 Batch    5/10   train_loss = 1.896
Epoch  82 Batch    0/10   train_loss = 1.833
Epoch  82 Batch    5/10   train_loss = 1.848
Epoch  83 Batch    0/10   train_loss = 1.882
Epoch  83 Batch    5/10   train_loss = 2.084
Epoch  84 Batch    0/10   train_loss = 1.965
Epoch  84 Batch    5/10   train_loss = 1.910
Epoch  85 Batch    0/10   train_loss = 1.836
Epoch  85 Batch    5/10   train_loss = 1.810
Epoch  86 Batch    0/10   train_loss = 1.745
Epoch  86 Batch    5/10   train_loss = 1.732
Epoch  87 Batch    0/10   train_loss = 1.632
Epoch  87 Batch    5/10   train_loss = 1.641
Epoch  88 Batch    0/10   train_loss = 1.581
Epoch  88 Batch    5/10   train_loss = 1.605
Epoch  89 Batch    0/10   train_loss = 1.535
Epoch  89 Batch    5/10   train_loss = 1.554
Epoch  90 Batch    0/10   train_loss = 1.485
Epoch  90 Batch    5/10   train_loss = 1.506
Epoch  91 Batch    0/10   train_loss = 1.441
Epoch  91 Batch    5/10   train_loss = 1.463
Epoch  92 Batch    0/10   train_loss = 1.397
Epoch  92 Batch    5/10   train_loss = 1.419
Epoch  93 Batch    0/10   train_loss = 1.362
Epoch  93 Batch    5/10   train_loss = 1.389
Epoch  94 Batch    0/10   train_loss = 1.342
Epoch  94 Batch    5/10   train_loss = 1.370
Epoch  95 Batch    0/10   train_loss = 1.306
Epoch  95 Batch    5/10   train_loss = 1.324
Epoch  96 Batch    0/10   train_loss = 1.295
Epoch  96 Batch    5/10   train_loss = 1.362
Epoch  97 Batch    0/10   train_loss = 1.297
Epoch  97 Batch    5/10   train_loss = 1.281
Epoch  98 Batch    0/10   train_loss = 1.225
Epoch  98 Batch    5/10   train_loss = 1.268
Epoch  99 Batch    0/10   train_loss = 1.189
Epoch  99 Batch    5/10   train_loss = 1.180
Epoch 100 Batch    0/10   train_loss = 1.135
Epoch 100 Batch    5/10   train_loss = 1.155
Epoch 101 Batch    0/10   train_loss = 1.106
Epoch 101 Batch    5/10   train_loss = 1.116
Epoch 102 Batch    0/10   train_loss = 1.064
Epoch 102 Batch    5/10   train_loss = 1.080
Epoch 103 Batch    0/10   train_loss = 1.047
Epoch 103 Batch    5/10   train_loss = 1.063
Epoch 104 Batch    0/10   train_loss = 1.004
Epoch 104 Batch    5/10   train_loss = 1.020
Epoch 105 Batch    0/10   train_loss = 0.986
Epoch 105 Batch    5/10   train_loss = 0.997
Epoch 106 Batch    0/10   train_loss = 0.945
Epoch 106 Batch    5/10   train_loss = 0.950
Epoch 107 Batch    0/10   train_loss = 0.907
Epoch 107 Batch    5/10   train_loss = 0.909
Epoch 108 Batch    0/10   train_loss = 0.863
Epoch 108 Batch    5/10   train_loss = 0.879
Epoch 109 Batch    0/10   train_loss = 0.836
Epoch 109 Batch    5/10   train_loss = 0.850
Epoch 110 Batch    0/10   train_loss = 0.814
Epoch 110 Batch    5/10   train_loss = 0.819
Epoch 111 Batch    0/10   train_loss = 0.786
Epoch 111 Batch    5/10   train_loss = 0.790
Epoch 112 Batch    0/10   train_loss = 0.764
Epoch 112 Batch    5/10   train_loss = 0.775
Epoch 113 Batch    0/10   train_loss = 0.735
Epoch 113 Batch    5/10   train_loss = 0.753
Epoch 114 Batch    0/10   train_loss = 0.725
Epoch 114 Batch    5/10   train_loss = 0.718
Epoch 115 Batch    0/10   train_loss = 0.681
Epoch 115 Batch    5/10   train_loss = 0.698
Epoch 116 Batch    0/10   train_loss = 0.707
Epoch 116 Batch    5/10   train_loss = 0.738
Epoch 117 Batch    0/10   train_loss = 0.701
Epoch 117 Batch    5/10   train_loss = 0.668
Epoch 118 Batch    0/10   train_loss = 0.643
Epoch 118 Batch    5/10   train_loss = 0.679
Epoch 119 Batch    0/10   train_loss = 0.647
Epoch 119 Batch    5/10   train_loss = 0.644
Epoch 120 Batch    0/10   train_loss = 0.626
Epoch 120 Batch    5/10   train_loss = 0.621
Epoch 121 Batch    0/10   train_loss = 0.592
Epoch 121 Batch    5/10   train_loss = 0.596
Epoch 122 Batch    0/10   train_loss = 0.579
Epoch 122 Batch    5/10   train_loss = 0.596
Epoch 123 Batch    0/10   train_loss = 0.620
Epoch 123 Batch    5/10   train_loss = 0.619
Epoch 124 Batch    0/10   train_loss = 0.593
Epoch 124 Batch    5/10   train_loss = 0.585
Epoch 125 Batch    0/10   train_loss = 0.550
Epoch 125 Batch    5/10   train_loss = 0.536
Epoch 126 Batch    0/10   train_loss = 0.530
Epoch 126 Batch    5/10   train_loss = 0.538
Epoch 127 Batch    0/10   train_loss = 0.517
Epoch 127 Batch    5/10   train_loss = 0.541
Epoch 128 Batch    0/10   train_loss = 0.551
Epoch 128 Batch    5/10   train_loss = 0.581
Epoch 129 Batch    0/10   train_loss = 0.526
Epoch 129 Batch    5/10   train_loss = 0.446
Epoch 130 Batch    0/10   train_loss = 0.422
Epoch 130 Batch    5/10   train_loss = 0.447
Epoch 131 Batch    0/10   train_loss = 0.443
Epoch 131 Batch    5/10   train_loss = 0.444
Epoch 132 Batch    0/10   train_loss = 0.416
Epoch 132 Batch    5/10   train_loss = 0.388
Epoch 133 Batch    0/10   train_loss = 0.361
Epoch 133 Batch    5/10   train_loss = 0.372
Epoch 134 Batch    0/10   train_loss = 0.395
Epoch 134 Batch    5/10   train_loss = 0.431
Epoch 135 Batch    0/10   train_loss = 0.416
Epoch 135 Batch    5/10   train_loss = 0.366
Epoch 136 Batch    0/10   train_loss = 0.326
Epoch 136 Batch    5/10   train_loss = 0.317
Epoch 137 Batch    0/10   train_loss = 0.310
Epoch 137 Batch    5/10   train_loss = 0.306
Epoch 138 Batch    0/10   train_loss = 0.295
Epoch 138 Batch    5/10   train_loss = 0.278
Epoch 139 Batch    0/10   train_loss = 0.253
Epoch 139 Batch    5/10   train_loss = 0.251
Epoch 140 Batch    0/10   train_loss = 0.253
Epoch 140 Batch    5/10   train_loss = 0.256
Epoch 141 Batch    0/10   train_loss = 0.247
Epoch 141 Batch    5/10   train_loss = 0.243
Epoch 142 Batch    0/10   train_loss = 0.235
Epoch 142 Batch    5/10   train_loss = 0.237
Epoch 143 Batch    0/10   train_loss = 0.235
Epoch 143 Batch    5/10   train_loss = 0.247
Epoch 144 Batch    0/10   train_loss = 0.257
Epoch 144 Batch    5/10   train_loss = 0.249
Epoch 145 Batch    0/10   train_loss = 0.228
Epoch 145 Batch    5/10   train_loss = 0.223
Epoch 146 Batch    0/10   train_loss = 0.238
Epoch 146 Batch    5/10   train_loss = 0.265
Epoch 147 Batch    0/10   train_loss = 0.288
Epoch 147 Batch    5/10   train_loss = 0.296
Epoch 148 Batch    0/10   train_loss = 0.281
Epoch 148 Batch    5/10   train_loss = 0.251
Epoch 149 Batch    0/10   train_loss = 0.232
Epoch 149 Batch    5/10   train_loss = 0.220
Epoch 150 Batch    0/10   train_loss = 0.202
Epoch 150 Batch    5/10   train_loss = 0.233
Epoch 151 Batch    0/10   train_loss = 0.230
Epoch 151 Batch    5/10   train_loss = 0.215
Epoch 152 Batch    0/10   train_loss = 0.190
Epoch 152 Batch    5/10   train_loss = 0.173
Epoch 153 Batch    0/10   train_loss = 0.164
Epoch 153 Batch    5/10   train_loss = 0.161
Epoch 154 Batch    0/10   train_loss = 0.156
Epoch 154 Batch    5/10   train_loss = 0.183
Epoch 155 Batch    0/10   train_loss = 0.238
Epoch 155 Batch    5/10   train_loss = 0.283
Epoch 156 Batch    0/10   train_loss = 0.233
Epoch 156 Batch    5/10   train_loss = 0.180
Epoch 157 Batch    0/10   train_loss = 0.163
Epoch 157 Batch    5/10   train_loss = 0.154
Epoch 158 Batch    0/10   train_loss = 0.149
Epoch 158 Batch    5/10   train_loss = 0.144
Epoch 159 Batch    0/10   train_loss = 0.128
Epoch 159 Batch    5/10   train_loss = 0.127
Epoch 160 Batch    0/10   train_loss = 0.119
Epoch 160 Batch    5/10   train_loss = 0.123
Epoch 161 Batch    0/10   train_loss = 0.114
Epoch 161 Batch    5/10   train_loss = 0.119
Epoch 162 Batch    0/10   train_loss = 0.111
Epoch 162 Batch    5/10   train_loss = 0.116
Epoch 163 Batch    0/10   train_loss = 0.108
Epoch 163 Batch    5/10   train_loss = 0.113
Epoch 164 Batch    0/10   train_loss = 0.105
Epoch 164 Batch    5/10   train_loss = 0.110
Epoch 165 Batch    0/10   train_loss = 0.102
Epoch 165 Batch    5/10   train_loss = 0.107
Epoch 166 Batch    0/10   train_loss = 0.100
Epoch 166 Batch    5/10   train_loss = 0.106
Epoch 167 Batch    0/10   train_loss = 0.100
Epoch 167 Batch    5/10   train_loss = 0.110
Epoch 168 Batch    0/10   train_loss = 0.109
Epoch 168 Batch    5/10   train_loss = 0.122
Epoch 169 Batch    0/10   train_loss = 0.121
Epoch 169 Batch    5/10   train_loss = 0.123
Epoch 170 Batch    0/10   train_loss = 0.106
Epoch 170 Batch    5/10   train_loss = 0.103
Epoch 171 Batch    0/10   train_loss = 0.094
Epoch 171 Batch    5/10   train_loss = 0.113
Epoch 172 Batch    0/10   train_loss = 0.132
Epoch 172 Batch    5/10   train_loss = 0.119
Epoch 173 Batch    0/10   train_loss = 0.091
Epoch 173 Batch    5/10   train_loss = 0.099
Epoch 174 Batch    0/10   train_loss = 0.089
Epoch 174 Batch    5/10   train_loss = 0.095
Epoch 175 Batch    0/10   train_loss = 0.087
Epoch 175 Batch    5/10   train_loss = 0.092
Epoch 176 Batch    0/10   train_loss = 0.084
Epoch 176 Batch    5/10   train_loss = 0.091
Epoch 177 Batch    0/10   train_loss = 0.083
Epoch 177 Batch    5/10   train_loss = 0.089
Epoch 178 Batch    0/10   train_loss = 0.081
Epoch 178 Batch    5/10   train_loss = 0.088
Epoch 179 Batch    0/10   train_loss = 0.080
Epoch 179 Batch    5/10   train_loss = 0.087
Epoch 180 Batch    0/10   train_loss = 0.079
Epoch 180 Batch    5/10   train_loss = 0.086
Epoch 181 Batch    0/10   train_loss = 0.078
Epoch 181 Batch    5/10   train_loss = 0.085
Epoch 182 Batch    0/10   train_loss = 0.077
Epoch 182 Batch    5/10   train_loss = 0.084
Epoch 183 Batch    0/10   train_loss = 0.076
Epoch 183 Batch    5/10   train_loss = 0.083
Epoch 184 Batch    0/10   train_loss = 0.075
Epoch 184 Batch    5/10   train_loss = 0.082
Epoch 185 Batch    0/10   train_loss = 0.075
Epoch 185 Batch    5/10   train_loss = 0.081
Epoch 186 Batch    0/10   train_loss = 0.074
Epoch 186 Batch    5/10   train_loss = 0.080
Epoch 187 Batch    0/10   train_loss = 0.073
Epoch 187 Batch    5/10   train_loss = 0.079
Epoch 188 Batch    0/10   train_loss = 0.072
Epoch 188 Batch    5/10   train_loss = 0.079
Epoch 189 Batch    0/10   train_loss = 0.072
Epoch 189 Batch    5/10   train_loss = 0.078
Epoch 190 Batch    0/10   train_loss = 0.071
Epoch 190 Batch    5/10   train_loss = 0.077
Epoch 191 Batch    0/10   train_loss = 0.070
Epoch 191 Batch    5/10   train_loss = 0.077
Epoch 192 Batch    0/10   train_loss = 0.070
Epoch 192 Batch    5/10   train_loss = 0.076
Epoch 193 Batch    0/10   train_loss = 0.069
Epoch 193 Batch    5/10   train_loss = 0.075
Epoch 194 Batch    0/10   train_loss = 0.068
Epoch 194 Batch    5/10   train_loss = 0.075
Epoch 195 Batch    0/10   train_loss = 0.068
Epoch 195 Batch    5/10   train_loss = 0.074
Epoch 196 Batch    0/10   train_loss = 0.067
Epoch 196 Batch    5/10   train_loss = 0.074
Epoch 197 Batch    0/10   train_loss = 0.067
Epoch 197 Batch    5/10   train_loss = 0.073
Epoch 198 Batch    0/10   train_loss = 0.066
Epoch 198 Batch    5/10   train_loss = 0.073
Epoch 199 Batch    0/10   train_loss = 0.066
Epoch 199 Batch    5/10   train_loss = 0.072
Model Trained and Saved

Save Parameters

Save seq_length and save_dir for generating a new TV script.


In [17]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Save parameters for checkpoint
helper.save_params((seq_length, save_dir))

Checkpoint


In [18]:
"""
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 [19]:
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)
    """
    # TODO: Implement Function
    input_ = loaded_graph.get_tensor_by_name("input:0")
    initial_state_ = loaded_graph.get_tensor_by_name("initial_state:0")
    final_state_ = loaded_graph.get_tensor_by_name("final_state:0")
    probs_ = loaded_graph.get_tensor_by_name("probs:0")
    return input_, initial_state_, final_state_, probs_


"""
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 [20]:
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
    """
    # TODO: Implement Function
    choice = np.random.choice([int(x) for x in int_to_vocab.keys()], p=probabilities)
    return int_to_vocab[choice]


"""
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 [21]:
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: power off, einstein.
moe_szyslak: so whaddaya want here? uh, appendectomy, lipo, or(calling, cool) ah," enemies the" saint" admirer for".
moe_szyslak: ah-ha, four old beautiful midge! i've been very good with my thumb.
barney_gumble: i understand. it can really lisa?
marge_simpson: like there's a good question?...
homer_simpson: hiya, quick.
barney_gumble: you stole my back everyday!
lenny_leonard: you know, i used to keep artie) the pity anniversary.--(looks his head)
lisa_simpson:(big wave) dad! have a hurt!
homer_simpson:(being guys) don't my best friend?
moe_szyslak:(relieved), it's so what?
can i don't think he'll say, coffee would you don't wanna jebediah old.
barney_gumble:(arguing to switch sound) there's your woulda touch the best / come on tonight!
warm_female_voice:... _montgomery_burns: the way to hold?
lenny_leonard: har, handsome!
carl_carlson: people burns like when

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.